Generalized free wreath products and their operator algebras

Define the f.n.s. \omega=\mu\circ E_N\in M_* and note that, for all t\in\mathbb{R} and x\in M, \omega\circ\tau_t(x)=\omega(x). Indeed, the equality is clear when x is reduced and for x\in N it follows from the hypothesis \mu\circ\gamma_t=\mu for all t\in\mathbb{R}. Define X:=\{x\in M\,:\,(t\mapsto\tau_t(x))\text{ is strongly continuous}\} and note that X is a vector subspace, stable by involution and contains M_1\cup M_2. Let x,y\in X and let us show that xy\in X. View M\subset\mathcal{L}({\rm L}^2(M,\omega)), where \xi_\omega is the cyclic vector in the GNS construction {\rm L}^2(M,\omega) of \omega. One has, for all s,t\in\mathbb{R}, \tau_t(xy)-\tau_s(xy)=\tau_t(x)(\tau_t(y)-\tau_s(y))+(\tau_t(x)-\tau_s(y))\tau_s(y). Hence, for all \xi\in H, and all sequence (t_n)_n converging to t one has:

since both x,y\in X. It follows that X is a subalgebra of M containing both M_1 and M_2. Hence, it suffices to show that X is strong* closed in M. Let (x_n)_n\in X be a sequence converging strongly to x\in M (in particular, (x_n) is norm bounded). It suffices to show that the set

is equal to {\rm L}^2(M,\omega). Let \xi=a\xi_\omega, where a\in M is analytic with respect to the modular group (\sigma_t)_t of \omega. Denoting by J the anti-unitary modular operator of \omega we have,

It follows that the sequence of continuous maps (t\mapsto\sigma_t(x_n)\xi)_n converges uniformly on \mathbb{R} to the map t\mapsto\sigma_t(x)\xi. Hence, a\xi_\omega\in Z for all a\in M analytic with respect to (\sigma_t)_t. Since this set is dense in {\rm L}^2(M,\omega), it suffices to show that Z is norm-closed in {\rm L}^2(M,\omega). Let \xi_n\in Z be a sequence converging to \xi\in{\rm L}^2(M,\omega). Writing \Vert\tau_t(x)\xi_n-\tau_t(x)\xi\Vert\leq\Vert x\Vert\Vert\xi_n-\xi\Vert, we can argue as before to deduce that \xi\in Z. It concludes the proof.

(1). Since ({\rm id}\otimes h)(u) is the orthogonal projection onto {\rm Mor}(1,u)=\mathbb{C}\eta and \Vert\eta\Vert^2=\psi(1)=1 one has ({\rm id}\otimes h)(u)=\eta\eta^\star. Writing \eta=\sum_{\kappa,i}\lambda_{\kappa,i}^{\frac{1}{2}}b^\kappa_{ii} we have

(2). Recall from Remark 2.8 that the pair (m^\star\eta,m^\star\eta) solves the conjugate equation for (u,u). Let us show, under the hypothesis of ergodicity, that the pair (m^\star\eta,m^\star\eta) is standard. It suffices to show that, for all T\in{\rm End}(u), \varphi_u(T)=\psi_u(T), where \varphi_u(T):=\eta^\star m({\rm id}_B\otimes T)m^\star\eta and \psi_u(T):=\eta^\star m(T\otimes {\rm id}_B)m^\star\eta. Note that both m(T\otimes{\rm id}_B)m^\star\eta and m({\rm id}_B\otimes T)m^\star\eta are in \mathrm{Mor}(1,u). Since the action is ergodic, there exist \lambda,\mu\in\mathbb{C} such that m(T\otimes{\rm id}_B)m^\star\eta=\lambda\eta and m({\rm id}_B\otimes T)m^\star\eta=\mu\eta. It suffices to show that \lambda=\mu. Write, in the basis of matrix units (e^\kappa_{ij})_{\kappa,i,j} in \mathcal{L}(B) such that Q_\kappa e^{\kappa}_{ij}=\lambda_{\kappa,i}e^{\kappa}_{ij}, T(e^\kappa_{ij})=\sum_{\zeta,k,l}T^{ij,\kappa}_{kl,\zeta}e^\zeta_{kl}, where T^{ij,\kappa}_{kl,\zeta}\in\mathbb{C} and note that m^\star\eta=\sum_{\kappa,i,j}\lambda_{\kappa,j}^{-1}e^\kappa_{ij}\otimes e^\kappa_{ji}. A direct computation gives:

Since m(T\otimes{\rm id}_B)m^\star\eta=\lambda \sum_{\kappa,i}e^\kappa_{ii} we find that the map \left[(\kappa,i)\mapsto\sum_k\lambda_{\kappa,k}^{-1}T^{ik,\kappa}_{ik,\kappa}\right] is constant, equals to \lambda. Similarly, the map \left[(\kappa,i)\mapsto\sum_k\lambda_{\kappa,k}^{-1}T^{ki,\kappa}_{ki,\kappa}\right] is constant, equals to \mu. Hence, for all 1\leq\kappa\leq K and all 1\leq k\leq N_\kappa, one has \mu=\lambda_{\kappa,1}^{-1}T^{1k,\kappa}_{1k,\kappa}+\sum_{i=2}^{N_\kappa}\lambda_{\kappa,i}^{-1}T^{ik,\kappa}_{ik,\kappa} which implies that T^{1k,\kappa}_{1k,\kappa}=\lambda_{\kappa,1}(\mu-\sum_{i=2}^{N_\kappa}\lambda_{\kappa,i}^{-1} T^{ik,\kappa}_{ik,\kappa}). It follows that:

We deduce that \lambda=\mu so (m^\star\eta,m^\star\eta) is standard. It follows that Q_u=J_u^*J_u, where \langle J_u(a),b\rangle=\langle m^\star\eta,a\otimes b\rangle=\langle 1_B,ab\rangle=\psi(ab)=\langle a^*,b\rangle for all a,b\in B. Hence, J_u(a)=a^*=S_\psi(a), the modular operator of \psi. It follows, by definition, that Q_u=S_\psi^*S_\psi=\nabla_\psi. Write S_\psi=J_\psi\nabla^{\frac{1}{2}}_\psi. A direct computation shows that J_\psi is the unique anti-unitary on B such that J_\psi(b^\gamma_{kl})=b^\gamma_{lk} and \nabla_\psi(b^\gamma_{kl})=\frac{\lambda_{\gamma,k}}{\lambda_{\gamma,l}}b^\gamma_{kl}. Hence, {\rm dim}_q(u)={\rm Tr}(\nabla_\psi)=\sum_\kappa{\rm Tr}(Q_\kappa){\rm Tr}(Q_\kappa^{-1})=\delta\sum_\kappa{\rm Tr}(Q_\kappa)=\delta. Since ({\rm id}\otimes\sigma_t^h)(u)=(\nabla_\psi^{it}\otimes 1)u(\nabla_\psi^{it}\otimes 1), it follows that ({\rm id}\otimes\sigma^h_t)(u)=\sum\left(\frac{\lambda_{\gamma,k}\lambda_{\kappa,i}}{\lambda_{\gamma,l}\lambda_{\kappa,j}}\right)^{it}b^\kappa_{ij}(b^\gamma_{kl})^\star\otimes u^{ij,\kappa}_{kl,\gamma} which proves the first part of (b). Moreover, since \sigma_t^\psi(b)=\nabla_\psi^{it}b\nabla_\psi^{-it} for all b\in B we find \sigma_t^\psi(b^\gamma_{kl})=\left(\frac{\lambda_{\gamma,k}}{\lambda_{\gamma,l}}\right)^{it}b^\gamma_{kl} from which we can easily check the last part of (b). (c) is proven the same way using the relation:

(1). The unique \beta-invariant state \psi is defined by ({\rm id}\otimes h)\beta=\psi(\cdot)1. Applying \chi_\kappa gives (1). Since \psi is faithful, the maps \beta_\kappa also are.

(2). Follows from Assertion (2.b) and (2.c) in Proposition 2.11.

(3). Since \sigma_t^{\psi_\kappa^{-1}}=\sigma_{-t}^{\psi_\kappa} it follows from (2) that the subalgebra \beta_\kappa(B)\subset M_{N_\kappa}(\mathbb{C})\otimes {\rm L}^\infty(H) is globally invariant under the modular group of \psi_\kappa^{-1}\otimes h. By Takesaki’s Theorem there exists a unique \psi_\kappa^{-1}\otimes h-preserving conditional expectation \mathbb{E}_\kappa\,:\,M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H)\rightarrow\beta_\kappa(B), which is normal and faithful. Then, E_\kappa:=\beta_\kappa^{-1}\circ \mathbb{E}_\kappa satisfies E_\kappa\circ\beta_\kappa={\rm id}_B and, using (1) and the fact that \mathbb{E}_\kappa is state preserving, we deduce that \psi\circ E_\kappa=\psi_\kappa^{-1}\otimes h. From this property and since \psi_\kappa^{-1}\otimes h is faithful, we deduce that E_\kappa is faithful. By (1), (\beta_\kappa(b^\gamma_{kl}))_{\gamma,k,l} is a \psi_\kappa^{-1}\otimes h-orthonormal basis of the finite dimensional subalgebra \beta_\kappa(B) which easily implies that

Using the formulas (b^\gamma_{kl})^*=\left(\frac{\lambda_{\gamma,k}}{\lambda_{\gamma,l}}\right)^{\frac{1}{2}}b^\gamma_{lk} and b^\kappa_{ip}b^\kappa_{qj}=\lambda_{\kappa,p}^{-\frac{1}{2}}\delta_{pq}b^\kappa_{ij} we get:

Since \psi_\kappa^{-1}(b^\kappa_{is})={\rm Tr}(Q_\kappa^{-1})^{-1}\lambda_{\kappa,s}^{-\frac{3}{2}}\delta_{is} and \left(u^{rs,\kappa}_{kl,\gamma}\right)^*=\left(\frac{\lambda_{\kappa,s}\lambda_{\gamma,k}}{\lambda_{\kappa,r}\lambda_{\gamma,l}}\right)^{\frac{1}{2}}u^{sr,\kappa}_{lk,\gamma} we deduce that:

(1). Fix x\in {\rm Pol}(H), 1\leq\kappa,\zeta,\gamma \leq K, 1\leq i,j\leq N_\kappa, 1\leq k,l\leq N_\gamma and 1\leq p,r\leq N_\zeta and define Z:=\sum_q\lambda_{\gamma,q}^{-1/2}(1\otimes u^{pr,\zeta}_{kq,\gamma})\Delta(u^{ij,\kappa}_{ql,\gamma}x). Using the relation

Hence, using invariance of the Haar state, and the previous equation we get:

(2). Using the formula in Proposition 2.13, we get:

Hence, for all \gamma,k,l we have \sum_{i,j}\lambda_{\kappa,i}^{-\frac{1}{2}}\lambda_{\kappa,j}^{-\frac{3}{2}}h(u^{ji,\kappa}_{kl,\gamma}x_{ij})=0. Using (1) we find, for all \gamma,\zeta,k,p,

Multiplying by \lambda_{\gamma,k}^{1/2} and summing over k,\gamma we get:

This concludes the proof.

The action \beta being 2-ergodic, it is ergodic so its unique invariant state \psi is a \delta-form. Moreover, ({\rm id}\otimes h)(u\otimes u) is the orthogonal projection onto \mathrm{Mor}(1,u\otimes u) = \mathbb{C} \eta\otimes \eta\oplus \mathbb{C} m^\star\eta. Write \eta=\sum_{\kappa,i}\lambda_{\kappa,i}^{\frac{1}{2}}b^\kappa_{ii}, m^\star\eta = \sum_{\kappa,i,t}\lambda_{\kappa,i}^{\frac{1}{2}}\lambda_{\kappa,t}^{-\frac{1}{2}} b^\kappa_{it}\otimes b^\kappa_{ti}, and therefore \Vert m^\star\eta \Vert^2 = \sum_\kappa {\rm Tr}(Q_\kappa^{-1}){\rm Tr}(Q_\kappa)= \sum_\kappa \delta {\rm Tr}(Q_\kappa) = \delta, because \psi is a \delta-form. Since (\eta^\star\otimes \eta^\star) m^\star\eta = 1 we have \Vert m^\star\eta - \eta \otimes \eta \Vert^2 = \Vert m^\star\eta \Vert^2-1 =: \delta-1. The orthogonal projection on the space is then the orthogonal projection that we obtain taking (\eta\otimes \eta, \frac{1}{\sqrt{\delta-1}}(m^\star\eta - \eta\otimes \eta)) as an orthogonal base. The projection is therefore given by p = \eta\eta^\star\otimes \eta\eta^\star + \frac{1}{\delta-1}(m^\star\eta - \eta\otimes \eta)(m^\star\eta - \eta\otimes \eta)^\star. Note that:

We have:

We deduce that the coefficient associated to (b_{ij}^\alpha \otimes b_{rs}^\kappa)\circ (b_{kl}^\gamma \otimes b_{vw}^\zeta)^\star is:

This coefficient is equal to h(u^{ij,\alpha}_{kl,\gamma}u^{rs,\kappa}_{vw,\zeta}).

(1). Let t\in\mathbb{R} and u\in\mathcal{U}(B\otimes C) be such that \sigma_{-t}^{\psi_\kappa}\otimes\sigma^h_t\otimes{\rm id}_C={\rm Ad}((\beta_\kappa\otimes{\rm id})(u)). Recall that \sigma_t^\psi={\rm Ad}(b_t), where b_t=\sum_{\gamma,r}\lambda_{\gamma,r}^{it}e^\gamma_{rr}=\sum_{\gamma,r}\lambda_{\gamma,r}^{it+1/2}b^\gamma_{rr}\in B is unitary and b_t^*=b_{-t}. Then, for all b\in B and c\in C,

Since \beta_\kappa is faithful, we deduce that (b_t\otimes 1)(b\otimes c)(b_{-t}\otimes 1)=u(b\otimes c)u^* for all b\in B, c\in C. Hence, v:=(b_{-t}\otimes 1)u\in\mathcal{Z}(B\otimes C). Since v is unitary and central in B\otimes C, we may write v=\sum_{\gamma,r}e^\gamma_{rr}\otimes\alpha_\gamma=\sum\lambda_{\gamma,r}^{1/2} b^\gamma_{rr}\otimes\alpha_\gamma where \alpha_\gamma\in \mathcal{U}(\mathcal{Z}(C)). Then, u=(b_t\otimes 1)v=\sum_{\gamma,r}\lambda_{\gamma,r}^{it+1/2} b^\gamma_{rr}\otimes \alpha_\gamma and u^*=\sum_{\zeta,s}\lambda_{\zeta,s}^{-it+1/2}b^\zeta_{ss}\otimes \alpha^*_\zeta. It follows that:

Fix 1\leq p\leq N_\kappa. Using the formula b^\kappa_{kt}b^\kappa_{pp}b^\kappa_{t'l}=\frac{\delta_{tp}\delta_{t'p}}{\lambda_{\kappa,p}}b^\kappa_{kl}, we find:

Note that (\sigma_{-t}^{\psi_\kappa}\otimes\sigma_t^h\otimes{\rm id}_C)(b^\kappa_{pp}\otimes 1\otimes 1)=b^\kappa_{pp}\otimes 1\otimes 1 hence,

By the Haar state formula given in Lemma 2.19 we have:

Applying h\otimes{\rm id}_C to Equation (\ref{EqT}) we get:

Define w:=\sum_{\gamma,r}\lambda_{\gamma,r}\lambda_{\gamma,r}^{it}\alpha_\gamma\in \mathcal{Z}(C). Note that, since N_\kappa\geq 2, \delta=\sum_{t}\lambda_{\kappa,t}^{-1}>\lambda_{\kappa,p}^{-1} hence,

and ww^*=\sum_{\gamma,\zeta,r,s}\left(\frac{\lambda_{\gamma,r}}{\lambda_{\zeta,s}}\right)^{it}\lambda_{\gamma,r}\lambda_{\zeta,s}\alpha_\gamma\alpha_{\zeta}^*=1. We conclude that w is unitary so the convex combination w=\sum_{\gamma,r}\lambda_{\gamma,r}\lambda_{\gamma,r}^{it}\alpha_\gamma is an extreme point in the unit ball of C which implies that \lambda_{\gamma,r}^{it}\alpha_\gamma=\lambda_{\zeta,s}^{it}\alpha_\zeta for all \gamma,\zeta,r,s. Writing a\in\mathcal{U}(\mathcal{Z}(C)) the unitary satisfying a=\lambda_{\gamma,r}^{it}\alpha_\gamma for all 1\leq\gamma\leq K and all 1\leq r\leq N_\gamma one has

Hence, for all x\in{\rm L}^\infty(H), one has:

It follows that \sigma_t^h={\rm id}.

(2) follows from (1) (with t=0).

(3) can be obtained through the same reasoning as for (1), replacing the modular group relation (\sigma_{-t}^{\psi_\kappa}\otimes \sigma_t^h)\beta_\kappa = \beta_\kappa\circ \sigma_t^\psi by the scaling group relation (\sigma_{-t}^{\psi_\kappa}\otimes \tau_t^H)\beta_\kappa = \beta_\kappa\circ \sigma_{-t}^\psi.

Since \beta is ergodic, \psi is a \delta-form. It is easy to check that the unique \delta-form on \mathbb{C}^K is the uniform normalized trace and \delta=K. Write u=\sum_{i,j=1}^Ke_{ij}\otimes u_{ij}, where e_{ij} are the canonical matrix units, the unitary representation of H associated with \beta. Note that \beta_i(\mathbb{C}^K)={\rm Span}\{u_{ij}\,:\,1\leq j\leq K\}. By Lemma 2.19 and since \delta=K we find:

Let E_i\,:\,{\rm L}^\infty(H)\rightarrow\mathbb{C}^K be the ucp map from Proposition 2.13. Using the formula above for the Haar state as well as the formula given in Proposition 2.13 we obtain:

Let i\neq j and w\in \beta_i(\mathbb{C}^K)\otimes C\cap \beta_j(\mathbb{C}^K)\otimes C be a unitary (i\neq j), then we can write

Applying E_i\otimes{\rm id}_C to w, we get that for any l, \nu_l = \frac{1}{K-1} \sum_{k\neq l} \mu_k. In particular, \frac{1}{K-1}\sum_{k\neq l} \mu_k is a unitary so it is an extreme point in the unit ball of C. It follows that \mu_k=\mu_{k'} for all k,k'\neq l. Since this holds for all l and K\geq 3, we deduce that \mu_k does not depend on k hence w is of the form w=1\otimes v, with v\in\mathcal{U}(C). The C*-algebra \beta_i(\mathbb{C}^K)\otimes C\cap\beta_j(\mathbb{C}^K)\otimes C being the linear span of its unitaries, we get \beta_i(\mathbb{C}^K)\otimes C\cap\beta_j(\mathbb{C}^K)\otimes C=\mathbb{C}1\otimes C.

The uniqueness being obvious, let us show the existence. Write \pi(a)=\sum_ie_i\otimes\pi_i(a) and note that, since \pi is a unital *-homomorphism, each \pi_i\,:\, C(G)\rightarrow A is a unital *-homomorphism. Define \pi_0\,:\, C(G)^{*N}*C(S_N^+)\rightarrow A by \pi_0\circ\nu_i:=\pi_i and \pi_0\vert_{C(S_N^+)}:=\iota. Note that the commutations relations [\pi(a),({\rm id}\otimes\iota)(\beta(b))]=0 imply that, for all a\in C(G) and all 1\leq i,j\leq N one has \pi_0(\nu_i(a)u_{ij}-u_{ij}\nu_i(a))=0. Hence I\subseteq\ker(\pi_0) which implies that there exists a unital *-homomorphism \widetilde{\pi}\,:\,C(G\wr_* S_N^+)\rightarrow A satisfying the properties of the Proposition.

(1). If B is finite dimensional, it is known [FMP24] that the universal quantum homomorphism \rho\,:\, C(G)\rightarrow B\otimes \mathcal{U} exists. Then, the C*-algebra

where I is the closed two sided ideal generated by

and by \{(\omega\otimes{\rm id})(\rho(f)-1_B\otimes f)\,:\,\omega\in B^*,\,f\in C(F)\} does the job.

(2). Suppose that C(\mathbb{G}) exists and define \mathcal{U}\subset C(\mathbb{G}) to be the C*-subalgebra of C(\mathbb{G}) generated by \{(\omega\otimes{\rm id})(\rho(a))\,:\,a\in C(G) ,\,\omega\in B^*\} so that \rho(C(G))\subset B\otimes\mathcal{U}. We show below that \rho\,:\, C(G)\rightarrow B\otimes\mathcal{U} is the universal quantum homomorphism from C(G) to B in the sense of [FMP24]. Since {\rm dim}(C(G))\geq 2 [FMP24, Theorem D] will imply that B is finite dimensional. Let \pi\,:\,C(G)\rightarrow B\otimes C be a unital *-homomorphism and define \iota'_H:=\varepsilon_H(\cdot)1_C\,:\,C(H)\rightarrow C. Note that ({\rm id}\otimes\iota'_H)(\beta(b))=b\otimes 1_C and since B is abelian we have [({\rm id}\otimes\iota'_H)(\beta(b)),\pi(a)]=0 for all a\in C(G) and b\in B. It follows that there exists a unique unital *-homomorphism \nu\,:\,C(\mathbb{G})\rightarrow C such that ({\rm id}\otimes\nu)\circ\rho=\pi (and \nu\circ\iota_H=\iota'_H). Defining \widetilde{\pi}:=\nu\vert_{\mathcal{U}} we have ({\rm id}\otimes\widetilde{\pi})\rho=\pi and the uniqueness of \widetilde{\pi} is clear.

(3). It is a direct consequence of the uniqueness in the universal property.

Let \iota':=\Delta_{H\times F}\,:\, C(H\times F)\rightarrow C(\mathbb{G})\otimes C(\mathbb{G}), from which we get \iota'_H=\Delta_H and \iota'_F=\Delta_F with commuting images and \rho':=(\beta_G\otimes{\rm id})({\rm id}\otimes\rho)\Delta_G\,:\,C(G)\rightarrow B\otimes C(\mathbb{G})\otimes C(\mathbb{G}). For a\in {\rm Pol}(G) one has \rho'(a)=\sum (\beta_G\otimes{\rm id})(a_{(1)}\otimes\rho(a_{(2)}))=\sum(\rho(a_{(1)})\otimes 1)(\beta\otimes{\rm id})(\rho(a_{(2)})), where \Delta_G(a)=\sum a_{(1)}\otimes a_{(2)}. Hence,

We also have that for f\in {\rm Pol}(F), \Delta_G(f)=\Delta_F(f)=\sum f_{(1)}\otimes f_{(2)}\in{\rm Pol}(F)\otimes{\rm Pol}(F) and:

The existence of \Delta\,:\, C(\mathbb{G})\rightarrow C(\mathbb{G})\otimes C(\mathbb{G}) now follows from the universal property.

Let us show that ({\rm id}_{\mathcal{L}(H_v\otimes B)}\otimes\Delta)(a(v))=a(v)_{12}a(v)_{13} \forall v\in\mathrm{Rep}(G).

Since u is a representation of H one has ({\rm id}_{\mathcal{L}(B)}\otimes\Delta)(u)=({\rm id}_{\mathcal{L}(B)}\otimes\Delta_H)(u)=u_{12}u_{13} hence,

On the other hand, viewing a(v)_{12}a(v)_{13} as a 4 legged object in \mathcal{L}(H_v)\otimes\mathcal{L}(B)\otimes C(\mathbb{G})\otimes C(\mathbb{G}), and using the relation u(L_B(b)\otimes 1)u^*=(L_B\otimes{\rm id})(\beta(b)), we get:

This shows that ({\rm id}_{\mathcal{L}(H_v\otimes B)}\otimes\Delta)(a(v))=a(v)_{12}a(v)_{13}, \forall v\in\mathrm{Rep}(G).

We now prove (6). Let v,w\in\mathrm{Rep}(G). Viewing (\Sigma_{23}\otimes 1)a(v)_{13}a(w)_{23}(\Sigma_{23}\otimes 1) as a 5 legged object in \mathcal{L}(H_v)\otimes\mathcal{L}(H_w)\otimes\mathcal{L}(B)\otimes\mathcal{L}(B)\otimes C(\mathbb{G}) one has:

Hence, for a,b\in B, we get, with Y:=(\Sigma_{23}\otimes 1)a(v)_{13}a(w)_{23}(1\otimes a\otimes 1\otimes b\otimes 1),

It follows that, for any S\in\mathrm{Mor}_G(v\otimes w,t) one has:

Since this is true for all a,b\in B, it follows that

To show that (C(\mathbb{G}),\Delta) is a CQG, we will use the version [FP16, Definition 1.1] of the definition of a CQG. We already checked the second condition of [FP16, Definition 1.1]. Let us now check the first condition. Let \mathcal{X} be the set of coefficients of a(v), for v\in{\rm Irr}(G). It suffices to show that the unital *-subalgebra of C(\mathbb{G}) generated by \mathcal{X} and {\rm Pol}(H) is dense in C(\mathbb{G}). For \omega\in B^* define \omega_0\in\mathcal{L}(B)^*, \omega_0(T):=\omega(T1_B). Since u(1_B\otimes1)=1_B\otimes 1 we have,

It follows that, for any v\in\mathrm{Rep}(G), (\omega\otimes{\rm id})\rho(v_{ij})\in\mathcal{X}, for any coefficient v_{ij} of v and any \omega\in B^*. Hence, \{(\omega\otimes{\rm id})\rho(a)\,:\,a\in{\rm Pol}(G),\,\omega\in B^*\}\subset{\rm Span}(\mathcal{X}). Moreover, since a(1)=u, \mathcal{X} also contains the coefficients of u. So the *-algebra generated by \mathcal{X} is dense in C(\mathbb{G}) whenever the action \beta is faithful. In general, the *-algebra generated by \mathcal{X} and {\rm Pol}(H) is dense in C(\mathbb{G}). It also implies (5), by the general theory of CQG, once we know that (C(\mathbb{G}),\Delta) is a CQG. To conclude the proof of (1) we only have to check the last condition of [FP16, Definition 1.1]. To do so, we use [FP16, Remark 1]. Let, for v\in\mathrm{Rep}(G), s_v\in\mathrm{Mor}_G(1,v\otimes\overline{v}) and s_{\overline{v}}\in\mathrm{Mor}_G(1,\overline{v}\otimes v) be a pair solving the conjugate equation for (v,\overline{v}). Since \eta\in\mathrm{Mor}_H(1,u), a(1)=u, assertion (6) gives:

Suppose that the pair (s_v,s_{\overline{v}}) is standard. To show that (C(\mathbb{G}),\Delta) is a CQG it remains to prove that S_{\overline{v}} is non-degenerated in the sense of [FP16, Remark 1]. Writing, as in Section ?, s_{\overline{v}}=\sum_iQ_{v,i}^{-\frac{1}{2}}e_i^{\overline{v}}\otimes e_i^{v} we find, as in the proof of [FP16, Proposition 2.7], and using the notations introduced at the end of Section ?,

From this formula, it is clear that S_{\overline{v}} is non-degenerated so (C(\mathbb{G}),\Delta) is a CQG. The same computation gives S_v=\sum_{\kappa,i,j,k}\sqrt{\frac{Q_{v,i}\lambda_{\kappa,j}}{\lambda_{\kappa,k}}}e^v_i\otimes b^\kappa_{jk}\otimes e^{\overline{v}}_i\otimes b^\kappa_{kj}. From these formulas, one checks easily that the pair (S_v,S_{\overline{v}}) solves the conjugate equation for (a(v),a(\overline{v})) so \overline{a(v)}\simeq a(\overline{v}) for all v\in\mathrm{Rep}(G).

Note that statement (4) is an easy consequence of the fact that \rho(f)=1_B\otimes f for all f\in C(F). Finally, the existence and uniqueness of the character \varepsilon_\mathbb{G} satisfying the properties given in (2) is a direct consequence of the universal property of C(\mathbb{G}). The fact that it is the counit of \mathbb{G} follows from the definition of \Delta and \varepsilon_\mathbb{G}.

Let us denote by \rho_0 the map defining \mathbb{G}_0:=G\wr_{*,\beta,F}H_0, \rho the one for \mathbb{G}:=G\wr_{*,\beta,F}H and view C(H_0\times F)\subset C(\mathbb{G}_0) and C(H\times F)\subset C(\mathbb{G}). Define \Sigma:=\mathbb{G}_0\underset{H_0 \times F}{\ast}(H\times F) and view C(\mathbb{G}_0),C(H\times F)\subset C(\Sigma). By the universal property of C(\mathbb{G}), there is a unique unital *-homomorphism \pi\,:\, C(\mathbb{G})\rightarrow C(\Sigma) such that:

It is easy to check that \pi intertwines the comultiplication. Moreover, the universal property of C(\mathbb{G}_0) gives a unique unital *-homomorphism \pi_0\,:\,C(\mathbb{G}_0)\rightarrow C(\mathbb{G}) such that ({\rm id}\otimes\pi_0)\rho_0=\rho and \pi_0\vert_{C(H_0\times F)} is the inclusion C(H_0\times F)\subset C(H\times F)\subset C(\mathbb{G}). Now, by the universal property of the amalgamated free product C(\Sigma)=C(\mathbb{G}_0)\underset{C(H_0\times F)}{*}C(H\times F), there is a unique unital *-homomorphism \pi^{-1}\,:\, C(\Sigma)\rightarrow C(\mathbb{G}) such that \pi^{-1}\vert_{C(\mathbb{G}_0)}=\pi_0 and \pi\vert_{C(H\times F)} is the inclusion C(H\times F)\subset C(\mathbb{G}). It is clear that \pi^{-1} is the inverse of \pi.

Let \mathbb{G}:=G\wr_{*,\beta,F}\Lambda. Recall that C(G^{*_F X}\rtimes\Lambda)=C(G^{*_F X})\otimes C(\Lambda) and define \rho\,:\,C(G)\rightarrow B^X\otimes C(G^{*_F X\times K})\subset B^X\otimes C(G^{*_F X\times K}\rtimes\Lambda) by \rho(a):=\sum_{x\in X,1\leq\kappa\leq K}e_{x,\kappa}\otimes\nu_{x,\kappa}(a) and both \iota_F\,:\,C(F)\rightarrow C(G^{*_F X\times K})\subset C(G^{*_F X\times K}\rtimes\Lambda) and \iota_\Lambda\,:\, C(\Lambda)\rightarrow C(G^{*_F X}\rtimes\Lambda) are the inclusions. It is clear that they satisfy the relations of C(\mathbb{G}). Moreover, with these maps, C(G^{*_F X\times F}\rtimes\Lambda) satisfies the universal property of C(\mathbb{G}). Indeed, if \pi\,:\, C(G)\rightarrow B^X\otimes C, \iota'_F\,:\, C(F)\rightarrow C and \iota'_\Lambda\,:\, C(\Lambda)\rightarrow C are unital *-homomorphisms such that

for all a\in C(G), b\in B^X, c\in C(\Lambda) and f\in C(F). Writing \pi(a)=\sum_{x,\kappa,i,j} e^{x,\kappa}_{ij}\otimes\pi_{ij}^{x,\kappa}(a) and applying (\chi_x\otimes {\rm id}) to the relation [\pi(a),({\rm id}\otimes\iota'_\Lambda)(\beta(b))]=0 we find, for all x\in X, a\in C(G) and b\in B^X,

Fix any \kappa, r,s and y\in X and apply this equation to b:=e^{\kappa,y}_{rs}\in B^X to get:

where u_{xy}:=1_{\{\gamma\in\Lambda\,:\,\gamma\cdot y=x\}}. This implies that, for all \kappa,i,r, all a\in C(G) and all x,y\in X,

Since 1=\sum_yu_{xy}\in C(\Lambda), for all x\in X, we deduce in particular that \pi^{x,\kappa}_{ir}(a)=\delta_{ir}\pi^{x,\kappa}_{ss}(a) for all x,\kappa,i,r,s,a. For a\in C(G) let \pi_{x,\kappa}(a)\in C be the element satisfying \pi_{x,\kappa}(a)=\pi^{x,\kappa}_{rr}(a) for all 1\leq r\leq N_\kappa and note that \pi is of the form:

Since the projections e_{x,\kappa} are pairwise orthogonal and \pi is a unital *-homomorphism, each \pi_{x,\kappa}\,:\,C(G)\rightarrow C is a unital *-homomorphism and, since \pi(f)=1_{B^X}\otimes\iota'_F(f), we have \pi_{x,\kappa}(f)=\iota'_F(f) for all f\in F. Moreover, Equation (\ref{EqCummutation}) also implies that, for all a\in C(G) and all x,y\in X and all 1\leq\kappa\leq K, \pi_{x,\kappa}(a)\iota'_\Lambda(u_{xy})=\iota'_\Lambda(u_{xy})\pi_{x,\kappa}(a). Hence, for all a\in C(G) and (x,\kappa)\in X\times K, \pi_{x,\kappa}(a) commutes with \iota'_\Lambda(c) for all c\in\mathcal{A}_x, where \mathcal{A}_x\subset C(\Lambda) is the unital *-algebra generated by \{u_{xy}\,:\, y\in X\}. Since the action \Lambda\curvearrowright X is free, \mathcal{A}_x separates the points of \Lambda and so \mathcal{A}_x=C(\Lambda), for all x\in X. This implies that for all a\in A, (x,\kappa)\in X\times K and c\in C(\Lambda) [\pi_{x,\kappa}(a),\iota'_\Lambda(c)]=0. Hence, there exists a unique unital *-homomorphism \widetilde{\pi}\,:\, C(G^{*_FX\times K})\otimes C(\Lambda)\rightarrow C such that \widetilde{\pi}\circ\nu_{x,\kappa}=\pi_{x,\kappa} and \widetilde{\pi}\circ\iota'_\Lambda=\iota_\Lambda. This shows the universal property hence we may and will assume that C(\mathbb{G})=C(G^{*_F X\times K})\otimes C(\Lambda) with maps \rho, \iota_\Lambda and \iota_F given above. Let us denote by \Delta_1 the comultiplication on C(\mathbb{G}) given by the free wreath product construction and by \Delta_2 the one given by the semi-direct product construction. It suffices to show that \Delta_1=\Delta_2. Since by construction, they both coincide with \Delta_\Lambda on C(\Lambda), it suffices to show that \Delta_1\circ\nu_{x,\kappa}=\Delta_2\circ\nu_{x,\kappa} for all (x,\kappa)\in X\times K. Note that, for all a\in{\rm Pol}(G), \Delta_2(\nu_{x,\kappa}(a))=\sum_{\gamma\in\Lambda}\nu_{x,\kappa}(a_{(1)})\iota_\Lambda(\delta_\gamma)\otimes\nu_{\gamma^{-1}\cdot x,\kappa}(a_{(2)}) and,

It follows that \Delta_1\circ\nu_{x,\kappa}=\Delta_2\circ\nu_{x,\kappa} for all (x,\kappa)\in X\times K.

The uniqueness is clear and the existence of \pi is a direct consequence of the universal property of C(G\wr_*{\rm Aut}^+(B,\psi)) and assertion (6) of Theorem 3.5. It is clear that \pi intertwines the comultiplications since a(u^x) is a unitary representation. Let us construct the inverse of \pi. By the universal property of C(G) and [FP16, Proposition 8.1] there exists a unique unital *-homomorphism \varphi_0\,:\,C(G)\rightarrow B\otimes C(G\wr_*{\rm Aut}^+(B,\psi)) such that

where r(v), for non irreducible v\in\mathrm{Rep}(G), is defined just before [FP16, Proposition 8.1] (denoted there a(v)). By the universal property of C({\rm Aut}^+(B,\psi)) there is a unique unital *-homomorphism \iota_B\,:\,C({\rm Aut}^+(B,\psi))\rightarrow C(G\wr_*{\rm Aut}^+(B,\psi)) such that ({\rm id}\otimes\iota_B)(u)=r(1). From [FP16, Proposition 8.1] we know that ({\rm id}_{H_v}\otimes m)\in\mathrm{Mor}_{G\wr_*{\rm Aut}^+(B,\psi)}(r(v)\otimes r(1),r(v)) and ({\rm id}_{H_v}\otimes m)\Sigma_{12}\in\mathrm{Mor}_{G\wr_*{\rm Aut}^+(B,\psi)}(r(1)\otimes r(v),r(v)). Hence,

Take an orthonormal basis (e_r)_r of H_v with matrix units (e_{rs})_{rs} in \mathcal{L}(H_v), write v=\sum_{r,s}e_{rs}\otimes v_{rs} and:

Note that \sum_{\gamma,r}\lambda_{\gamma,r}^{\frac{1}{2}}r(1)^{ij,\kappa}_{rr,\gamma}=\delta_{ij}\lambda_{\kappa,i}^{\frac{1}{2}}, since \iota_B is a unital *-homomorphism. From the relation b^\kappa_{ip}b^{\zeta}_{qj}=\delta_{\kappa\zeta}\delta_{pq}\lambda^{-\frac{1}{2}}_{\kappa,p}b^\kappa_{ij} we find m\circ (b^\kappa_{ip}\circ (b^\zeta_{kl})^\star\otimes b^{\kappa'}_{qj}\circ (b^{\zeta'}_{k'l'})^\star)=\delta_{\kappa,\kappa'}\delta_{p,q}\lambda_{\kappa,p}^{-1/2}b^\kappa_{ij}\circ ((b^\zeta_{kl})^\star\otimes (b^{\zeta'}_{k'l'})^\star).

It follows that ({\rm id}_{H_v}\otimes m\otimes 1)r(v)_{124}r(1)_{34} is equal to:

and ({\rm id}_{H_v}\otimes m\otimes 1)r(1)_{24}r(v)_{134} is equal to:

Then, for all r,s,\kappa,i,j,\zeta,k,l,\zeta',k',l' one has

Note that (b_{kl}^\zeta)^\star(1_B)=\psi((b_{kl}^\zeta)^*)=\delta_{kl}\lambda_{\zeta,k}^{1/2} hence,

It follows that \varphi_0(v_{rs})=\sum b_{ij}^\kappa\otimes \lambda_{\zeta,q}^{1/2} r(v)^{ij,\kappa}_{qq,\zeta,rs}. Writing \beta(b^\alpha_{kl})=\sum_{\kappa,i,j}b^\kappa_{ij}\otimes u^{ij,\kappa}_{kl,\alpha} and using Equation (\ref{Eqfwp1}) we deduce that, with X:=({\rm id}\otimes\iota_B)(\beta(b^\alpha_{kl}))\varphi_0(v_{rs}),

The same computation with Y:=\varphi_0(v_{rs})({\rm id}\otimes\iota_B)(\beta(b^\alpha_{kl})) gives:

Hence, ({\rm id}\otimes\iota_B)(\beta(b^\alpha_{kl}))\varphi_0(v_{rs})=\varphi_0(v_{rs})({\rm id}\otimes\iota_B)(\beta(b^\alpha_{kl})) and since this holds for any v\in\mathrm{Rep}(G) and any \zeta,k,l, we deduce that [\varphi_0(a),({\rm id}\otimes\iota_B)(\beta(b))]=0 for all a\in C(G) and all b\in B. By the universal property of C(G\wr_{*,\beta}{\rm Aut}^+(B,\psi)), there exists a unique unital *-homomorphism \varphi\,:\,C(G\wr_{*,\beta}{\rm Aut}^+(B,\psi))\rightarrow C(G\wr_{*}{\rm Aut}^+(B,\psi)) such that \varphi\vert_{C({\rm Aut}^+(B,\psi))}=\iota_B and ({\rm id}\otimes\varphi)\rho=\varphi_0.

View \beta_i\circ\chi_i\,:\, B\rightarrow B_i\otimes C(H_i)\subset B_i\otimes C(H). By the universal property of direct sum there exists a unique unital *-homomorphism \beta\,:\, B\rightarrow B\otimes C(H) such that (\chi_i\otimes{\rm id})\beta=\beta_i\circ\chi_i. Since H_i is a dual quantum subgroup of H one has, for all i\in I,

Hence, ({\rm id}\otimes\Delta_H)\beta=(\beta\otimes{\rm id})\beta. Note that (\chi_i\otimes\varepsilon_H)\beta=({\rm id}\otimes\varepsilon_H)\beta_i\circ\chi_i=({\rm id}\otimes\varepsilon_{H_i})\beta_i\circ\chi_i=\chi_i(\cdot)1 for all i\in I. Hence, ({\rm id}\otimes\varepsilon_H)\beta={\rm id}_B and \beta is an action. The uniqueness and the last statement are clear.

Write C(H)=\langle C(H_p),u_e\vert p\in V(\mathcal{G}),\,e\in E(\mathcal{G})\rangle. Let \mathbb{G}:=G\wr_{*,\beta,F}H, \mathbb{G}_p:=G\wr_{*,\beta_p,F}H_p with canonical homomorphisms \rho\,:\, C(G)\rightarrow B\otimes C(\mathbb{G}) and \rho_p\,:\, C(G)\rightarrow B_p\otimes C(\mathbb{G}_p) and view C(H\times F)\subset C(\mathbb{G}) and C(H_p\times F)\subset C(\mathbb{G}_p). Define \mathbb{G}':=\pi_1(\mathbb{G}_p,\Sigma_e\times F,\mathcal{G},\mathcal{T}) and write C(\mathbb{G}')=\langle C(\mathbb{G}_p), v_e\vert p\in V(\mathcal{G}),\,e\in E(\mathcal{G})\rangle. Note that, for all e\in E(\mathcal{G}) and b\in C(\Sigma_e) one has, viewing C(\Sigma_e)\subset C(\Sigma_e\times F), r'_e(b)=v_{\overline{e}}(s'_e(b))v_e. By the universal property of C(H) there exists a unique unital *-homomorphism \nu_{H}\,:\, C(\mathbb{G})\rightarrow C(\mathbb{G}') such that \nu_{H}(u_e)=v_e and \nu_{H}\vert_{C(H_p)} is the inclusion C(H_p)\subset C(\mathbb{G}_p)\subset C(\mathbb{G}').

Note that, by definition, all the maps s_e'\vert_{C(F)}\,:\, C(F)\subset C(\Sigma_e\times F)\rightarrow C(\mathbb{G}_{s(e)})\subset C(\mathbb{G}') coincide, for e\in E(\mathcal{G}). Write \nu_F\,:\, C(F)\rightarrow C(\mathbb{G}') the unique map such that \nu_F=s_e'\vert_{C(F)} for all e\in E(\mathcal{G}). The fundamental quantum group relations give, for all f\in C(F) and e\in E(\mathcal{G}),

Hence, \nu_F(C(F)) commutes with v_e for all e\in E(\mathcal{G}). Also, by the free wreath product relations, C(F)\subset C(\mathbb{G}_p) and C(H_p)\subset C(\mathbb{G}_p) commute, for all p\in V(\mathcal{G}). Hence, \nu_F(C(F)) commutes with the C*-algebra generated by C(H_p), p\in V(G) and v_e, e\in E(\mathcal{G}) inside C(\mathbb{G}'). This C*-algebra is exactly \nu_H(C(H)) so we have the relation [\nu_H(x),\nu_F(f)]=0 for all a\in C(H) and f\in C(F).

Viewing C(\mathbb{G}_p)\subset C(\mathbb{G}'), we may view each map \rho_p as a map C(G)\rightarrow B_p\otimes C(\mathbb{G}'). By universal property of the direct sum there exists a unique unital *-homomorphism \rho'\,:\, C(G)\rightarrow B\otimes C(\mathbb{G}') such that (\chi_p\otimes{\rm id})\rho'=\rho_p for all p\in V(\mathcal{G}), where \chi_p\,:\, B\rightarrow B_p is the canonical surjection. Note that, for all p\in V(\mathcal{G}),

Hence, [\rho'(a),({\rm id}\otimes\nu_{H})(\beta(b))]=0 for all a\in C(G), b\in B. Moreover,

Hence, \rho'(f)=1_{B}\otimes f for all f\in C(F).

By the universal property of C(\mathbb{G}), there exists a unique unital *-homomorphism

Consider now the canonical maps:

We claim that:

Indeed, for a\in C(G) and b\in B one has,

Let us show that \Delta_{\mathbb{G}'}\circ\pi=(\pi\otimes\pi)\circ\Delta_{\mathbb{G}}. Since \nu_H \,:\, C(H)\rightarrow C(\mathbb{G}') intertwines the comultiplications and H is a dual subgroup of \mathbb{G}, it suffices to show that

Using Equation (\ref{EqBetaG}) we find, for all p\in V(\mathcal{G}),

This shows that \pi intertwines the comultiplications. It remains to show that \pi is invertible. To do so, we construct the inverse of \pi. Consider the unital *-homomorphism

and view C(H_p)\subset C(H)\subset C(\mathbb{G}) and C(F)\subset C(\mathbb{G}). It is clear that C(F) commutes with C(H_p), \rho'_p(f)=1_{B_p}\otimes f for f\in C(F) and [\rho_p'(a),\beta_p(b)]=0 for all a\in C(G) and b\in B_p. Hence there exists a unique unital *-homomorphism \pi_p^{-1}\,:\, C(\mathbb{G}_p)\rightarrow C(\mathbb{G}) such that ({\rm id}\otimes\pi_p^{-1})\rho_p=\rho_p' and \pi_p^{-1}\vert_{C(H_p)} is the inclusion of C(H_p) in C(\mathbb{G}). In particular, for each vertex p\in V(\mathcal{G}), \pi_p^{-1}\vert_{C(F)} is the inclusion of C(F) in C(\mathbb{G}). Consider the unitaries u_e\in C(H)\subset C(\mathbb{G}). By the universal property of the fundamental quantum group algebra, there exists a unique morphism \pi^{-1}\,:\, C(\mathbb{G}')\rightarrow C(\mathbb{G}) such that \pi^{-1}\vert_{C(\mathbb{G}_p)}=\pi_p^{-1} and \pi^{-1}(v_e)=u_e, for all p\in V(\mathcal{G}) and e\in E(\mathcal{G}). It is now easy to check that \pi^{-1} is the inverse of \pi.

If such D_\kappa is given with the corresponding *-morphisms we define

and \mathcal{D}:= D_\kappa\underset{C(H)\otimes_{{\rm max}} C(F)}{\ast}\mathcal{D}_0, where the amalgamation is with respect to the faithful unital *-homomorphisms \iota_\kappa\,:\,C(H)\otimes_{{\rm max}} C(F)\rightarrow D_\kappa and the inclusion C(H)\otimes_{{\rm max}} C(F)\rightarrow\mathcal{D}_0. Composing \pi_\kappa (resp. \rho_\gamma) with the inclusion of D_\kappa (resp. C_\gamma(\mathbb{G}) for \gamma\neq\kappa) in \mathcal{D} we may view \pi_\kappa\,:\,C(G)\rightarrow M_{N_\kappa}(\mathbb{C})\otimes \mathcal{D} (resp. \rho_\gamma\,:\,C(G)\rightarrow M_{N_\gamma}(\mathbb{C})\otimes\mathcal{D}). Use the universal property of the direct sum to get a map \rho' : C(G)\rightarrow B\otimes\mathcal{D} such that (\chi_\gamma \otimes {\rm id}) \rho' = \rho_\gamma for \gamma\neq \kappa and (\chi_\kappa \otimes {\rm id}) \rho' = \pi_\kappa, we also have a map \iota\,:\,C(H)\otimes_{{\rm max}} C(F)\rightarrow \mathcal{D} which is the canonical inclusion. Note that, if a\in C(G) and b\in B, the commutation relation \rho'(a)(({\rm id}\otimes \iota)(\beta(b)\otimes 1_F)) = (({\rm id}\otimes \iota)(\beta(b)\otimes 1_F))\rho'(a) can be checked when composed with any \chi_\gamma\otimes {\rm id}. For f\in C(F), the relation \rho'(f) = 1_B\otimes \iota(1\otimes f) also follows from the fact that we have the relations when applying (\chi_\gamma\otimes {\rm id}) and the fact that we amalgamated the different copies of C(H)\otimes_{{\rm max}} C(F). Applying the universal property of C(\mathbb{G}) gives a *-morphism \widetilde{\pi} : C(\mathbb{G})\rightarrow \mathcal{D}, such that \rho' = ({\rm id}\otimes\widetilde{\pi})\circ \rho and \widetilde{\pi}_{\vert C(H)\otimes_{{\rm max}} C(F)} =\iota. Define \widetilde{\pi}_\kappa :=\widetilde{\pi}_{\vert C_\kappa(\mathbb{G})}\,:\, C_\kappa(\mathbb{G}) \rightarrow \mathcal{D}. Note that \pi_\kappa=(\chi_\kappa\otimes{\rm id})\rho'=(\chi_\kappa\otimes \widetilde{\pi})\rho=({\rm id}\otimes\widetilde{\pi})\rho_\kappa=({\rm id}\otimes \widetilde{\pi}_\kappa)\rho_\kappa. In particular, \widetilde{\pi}_\kappa(C_\kappa(\mathbb{G}))\subset D_\kappa and we may view \widetilde{\pi}_\kappa as a map from C_\kappa(\mathbb{G}) to D_\kappa. Hence, \widetilde{\pi}_\kappa\vert_{C(H)\otimes_{{\rm max}} C(F)}=\iota_\kappa and this concludes the proof.

(1). The existence of \pi follows from the universal property of the amalgamated free product. Let us construct the inverse of \pi. Denote by C the free product of the C*-algebras C_\kappa(\mathbb{G}), amalgamated over C(H)\otimes_{{\rm max}} C(F) and view C(H)\otimes_{{\rm max}} C(F)\subset C. Note that we have \rho_\kappa\,:\,C(G)\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G})\subset M_{N_\kappa}(\mathbb{C})\otimes C such that \rho_\kappa(f)=1_\kappa\otimes f for all f\in C(F) and 1\leq\kappa\leq K. Hence, there exists a unique \rho'\,:\,C(G)\rightarrow B\otimes C such that (\chi_\kappa\otimes{\rm id})\rho'=\rho_\kappa for all \kappa and \rho'(f) = 1_B\otimes f for all f\in C(F). Since the images of \rho_\kappa and \beta_\kappa commute in M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G}) for all \kappa, it follows that the images of \rho' and \beta commute in B\otimes C. By the universal property of C(\mathbb{G}), there exists a unique \varphi\,:\,C(\mathbb{G})\rightarrow C such that \varphi\vert_{C(H)\otimes_{{\rm max}} C(F)} is the inclusion and ({\rm id}\otimes\varphi)\rho=\rho'. It is clear that \varphi is the inverse of \pi.

(2). During this proof we use the notation:

and we write \nu_\kappa\,:\, B\otimes C(G)\rightarrow \mathcal{A}_\kappa and \mu_\kappa\,:\, M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F)\rightarrow\mathcal{A}_\kappa the canonical faithful unital *-homomorphisms. They satisfy \mu_\kappa(\beta_\kappa(b)\otimes f)=\nu_\kappa(b\otimes f) for all b\in B and f\in C(F). We fix once and for all a basis of M_{N_\kappa}(\mathbb{C}) consisting of matrix units (e^\kappa_{ij})_{ij}.

The existence of \pi_\kappa is a consequence of the universal property of the amalgamated free product \mathcal{A}_\kappa. Let us construct its inverse \varphi_\kappa\,:\,M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G})\rightarrow \mathcal{A}_\kappa. We define the morphism \iota\,:\,C(H)\otimes_{{\rm max}} C(F)\rightarrow\mathcal{A}_\kappa, x\otimes f\mapsto\mu_\kappa(1\otimes x\otimes f) and we consider the elements

A direct computation gives the relations:

1. \varphi^\kappa_{ij}(ab)=\sum_k\varphi^\kappa_{ik}(a)\varphi^\kappa_{kj}(b).

2. (\varphi^\kappa_{ij}(a))^*=\varphi^\kappa_{ji}(a^*).

3. \varphi^\kappa_{ij}(1)=\delta_{ij}1_{\mathcal{A}_\kappa}.

4. \varphi^\kappa_{ij}(f)= \delta_{ij}\iota(1\otimes f) if f\in C(F).

Note that (ii), (iii) and (iv) are clear so we only prove (i). Let a,b\in C(G) then,

This relations mean that \varphi_\kappa\,:\, C(G)\rightarrow M_{N_\kappa}(\mathbb{C})\otimes \mathcal{A}_\kappa defined by \varphi_\kappa(a):=\sum_{i,j=1}^{N_\kappa}e^\kappa_{ij}\otimes\varphi^\kappa_{ij}(a) is a unital *-homomorphism. Let us check that \varphi_\kappa(a) and ({\rm id}\otimes\iota)(\beta_\kappa(b)\otimes 1) commutes, for all a\in C(G) and b\in B. We write \beta_\kappa\,:\,B\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C(H) with coefficients \beta_\kappa(b)=\sum_{i,j}e^\kappa_{ij}\otimes\beta^\kappa_{ij}(b). Then, for all b\in B,

Hence, writing ({\rm id}\otimes\iota)(\beta_\kappa(b)\otimes 1)\varphi_\kappa(a)=\sum_{i,j}e^\kappa_{ij}\otimes X^\kappa_{ij},

This shows that ({\rm id}\otimes\iota)(\beta_\kappa(b)\otimes 1)\varphi_\kappa(a)=\varphi_\kappa(a)({\rm id}\otimes\iota)(\beta_\kappa(b)\otimes 1) for all a\in C(G) and b\in B. Note that, for all f\in C(F), \varphi_\kappa(f) = \sum_{i,j} e_{ij}^\kappa \otimes \delta_{ij}\iota(1\otimes f) = 1_\kappa \otimes \iota(1\otimes f). The universal property of C_\kappa(\mathbb{G}) (Proposition 4.1) gives a unital *-homomorphism, still denoted \varphi_\kappa\,:\,C_\kappa(\mathbb{G})\rightarrow \mathcal{A}_\kappa such that \varphi_\kappa(\rho^\kappa_{ij}(a))=\varphi^\kappa_{ij}(a) for all 1\leq i,j\leq N_\kappa and all a\in C(G) and \varphi_\kappa(x\otimes f)=\mu_\kappa(1\otimes x\otimes f) for all x\in C(H) and f\in C(F). Now, we note that \varphi_\kappa(C_\kappa(\mathbb{G}))\subseteq\mu_\kappa(M_{N_\kappa}(\mathbb{C})\otimes 1\otimes 1)'\cap\mathcal{A}_\kappa. Indeed, it is clear that, for all x\in C(H) and f\in C(F), \varphi_\kappa(x\otimes f)=\mu_\kappa(1\otimes x\otimes f) commutes with \mu_\kappa(M_{N_\kappa}(\mathbb{C})\otimes 1\otimes 1) and, if x=\rho^\kappa_{ij}(a) is a coefficient of \rho_\kappa, then:

Hence, \varphi_\kappa(x) commutes with \mu_\kappa(M_{N_\kappa}(\mathbb{C})\otimes 1\otimes 1) for any x which is a coefficients of \rho_\kappa. Since C_\kappa(\mathbb{G}) is generated by C(H) and coefficients of \rho_\kappa, it follows that

Hence, there exists a unique unital *-homomorphism, still denoted \varphi_\kappa\,:\, M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G})\rightarrow \mathcal{A}_\kappa such that \varphi_\kappa(b\otimes x)=\mu_\kappa(b\otimes 1\otimes 1)\varphi_\kappa(x), for all b\in M_{N_\kappa}(\mathbb{C}) and x\in C_\kappa(\mathbb{G}). By construction we have \varphi_\kappa\circ\rho_\kappa(a)=\nu_\kappa(1\otimes a) for all a\in C(G) and \varphi_\kappa\vert_{M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F)}=\mu_\kappa. Since M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G}) is generated, as a C*-algebra, by \rho_\kappa(C(G))\cup M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F) this easily implies that \varphi_\kappa is the inverse of \pi_\kappa.

(1). It is easy to see that X_\kappa is a unital *-subalgebra of M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G}) containing M_{N_\kappa}(\mathbb{C})\otimes {\rm Pol}(H)\otimes {\rm Pol}(F). Note that it also contains 1\otimes\rho^\kappa_{ij}(a) for all a\in {\rm Pol}(G). Indeed, writing 1\otimes\rho^\kappa_{ij}(a)=\sum_r(e^\kappa_{ri}\otimes 1)\rho_\kappa(a)(e^\kappa_{jr}\otimes 1) and e^{\kappa}_{rs}=(e^\kappa_{rs})^\circ+\frac{\delta_{rs}}{\delta\lambda_{\kappa,r}}1_B, where E_\kappa((e^\kappa_{rs})^\circ\otimes 1)=0 we see that 1\otimes\rho^\kappa_{ij}(a)\in X when E_F(a)=0 and, for any a\in {\rm Pol}(G), writing a=a^\circ+E_F(a) we have \rho_{ij}^\kappa(a)=\rho^\kappa_{ij}(a^\circ)+\rho^\kappa_{ij}(E_F(a)) = \rho^\kappa_{ij}(a^\circ) + \delta_{ij}(1\otimes 1\otimes E_F(a)) which implies that 1\otimes\rho^\kappa_{ij}(a)\in X.

(2). As in the proof of Theorem 4.2, we write

where the amalgamation is with respect to the inclusion B\otimes C(F)\rightarrow B\otimes C(G) and the unital *-homomorphism \beta_\kappa\otimes{\rm id}_F\,:\, B\otimes C(F)\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F) and we write \nu_\kappa\,:\, B\otimes C(G)\rightarrow \mathcal{A}_\kappa and \mu_\kappa\,:\, M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F)\rightarrow\mathcal{A}_\kappa the canonical faithful unital *-homomorphism. Consider the unique ucp map E_2\,:\,\mathcal{A}_\kappa\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F) such that E_2\circ\mu_\kappa={\rm id}, E_2\circ\nu_\kappa=\beta_\kappa\otimes E_F and E_2(x)=0 for all x reduced of length \geq 2 in \mathcal{A}_\kappa in the amalgamated free product sense and with respect to the ucp maps ({\rm id}\otimes E_F)\,:\,B\otimes C(G)\rightarrow B\otimes C(F) and

Recall that \varphi_\kappa:=\pi_\kappa^{-1}\,:\, M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G})\rightarrow \mathcal{A}_\kappa is constructed in Theorem 4.2. Define the ucp map \mathbb{E}_\kappa:=E_2\circ\varphi_\kappa\,:\,M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G})\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F).

If x=X_0\rho_\kappa(a_1)X_1\dots \rho_\kappa(a_n)X_n (n\geq 1) is reduced then

is reduced in \mathcal{A}_\kappa of length at least 2 so E_2(y)=0. It follows that \mathbb{E}_\kappa(x)=E_2(y)=0. We prove in the same way that \mathbb{E}_\kappa(x)=0 for all types of reduced operators. Hence, \mathbb{E}_\kappa(\mathcal{C}_\kappa)=\{0\}. Recall that \varphi_\kappa\vert_{M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes C(F)}=\mu_\kappa and E_2\circ\mu_\kappa={\rm id}. Hence, \mathbb{E}_\kappa\vert_{M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F)}={\rm id}. It remains to show that \ker(\mathbb{E}_\kappa)\subset \overline{\mathcal{C}_\kappa} and this follows from (1) and the fact that \mathbb{E}_\kappa is the identity on the subalgebra M_{N_\kappa}(\mathbb{C})\otimes C(H)\otimes_{{\rm max}} C(F).

(3) By definition \Delta we have, writing \Delta_G(a)=\sum a_{(1)}\otimes a_{(2)}, with a_{(1)},a_{(2)}\in{\rm Pol}(G),

Since \Delta(C(H)\otimes_{{\rm max}} C(F))=\Delta_{H\times F}(C(H\times F))\subset (C(H)\otimes_{{\rm max}} C(F)) \otimes (C(H)\otimes_{{\rm max}} C(F)) we conclude easily that \Delta(C_\kappa(\mathbb{G}))\subset C_\kappa(\mathbb{G})\otimes C(\mathbb{G}).

Let x be a reduced operator and let us show that ({\rm id}\otimes\Delta)(x)\in\mathcal{C}_\kappa\otimes C(\mathbb{G}). We prove our statement for x a reduced operator of the form x=\rho_\kappa(a_1)X_1\dots \rho_\kappa(a_n)X_n, the proof for other types of reduced operator is the same.

Claim. If for all s, X_s\in M_{N_\kappa}(\mathbb{C})\otimes{\rm Pol}(H)\otimes\mathbb{C}1_{C(F)} then ({\rm id}\otimes\Delta)(x)\in\mathcal{C}_\kappa\otimes C(\mathbb{G}).

Proof of the Claim. Write x=\rho_\kappa(a_1)(X_1\otimes 1)\dots \rho_\kappa(a_n)(X_n\otimes 1), where X_s\in M_{N_\kappa}(\mathbb{C})\otimes{\rm Pol}(H) is such that E_\kappa(X_s)=0. Writing X_s\in M_{N_\kappa}(\mathbb{C})\otimes {\rm Pol}(H) as X_s=\sum_{i_s,j_s}b^\kappa_{i_sj_s}\otimes x_{i_sj_s} and ({\rm id}\otimes\Delta)(X_s\otimes 1)=({\rm id}\otimes\Delta_H)(X_s)\otimes 1_{C(F)}=\sum (X_s)_{(1)}\otimes (X_s)_{(2)} where

with \Delta_{H}(x_{i_sj_s})=\sum(x_{i_sj_s})_{(1)}\otimes (x_{i_sj_s})_{(2)}, we obtain:

Where \Delta_G(a_k)=\sum (a_k)_{(1)}\otimes (a_k)_{(2)} and E_F((a_k)_{(1)})=0. If (E_\kappa\otimes {\rm id}_F)((X_s)_{(1)})=0, for all s, then ({\rm id}\otimes\Delta)(x)\in \mathcal{C}_\kappa\otimes C(\mathbb{G}), since

Otherwise, we write (X_s)_{(1)}=(X_s)_{(1)}^\circ+(\beta_\kappa E_\kappa\otimes{\rm id}_F)((X_s)_{(1)}), where (E_\kappa\otimes{\rm id}_F)((X_s)_{(1)}^\circ)=0 so that we may decompose ({\rm id}\otimes\Delta)(x) as the sum of two terms Y_s+Z_s where in Y_s is the same as ({\rm id}\otimes\Delta)(x) but replacing (X_s)_{(1)} by (X_s)_{(1)}^\circ and the second term is:

By Proposition 2.13 we have:

Replacing this expression in Z_s we get:

Since E_\kappa(X_s)=0, we may now use Lemma 2.15 to deduce that Z_s=0. Doing that for all s, we conclude that ({\rm id}\otimes\Delta)(x)\in\mathcal{C}_\kappa\otimes C(\mathbb{G}).\hfill \qed

End of the proof of (3). Fix a basis (y_k)_k of {\rm Pol}(F). If now x is an arbitrary reduced operator of the form x=\rho_\kappa(a_1)X_1\dots \rho_\kappa(a_n)X_n, with X_s\in M_{N_\kappa}(\mathbb{C})\otimes {\rm Pol}(H)\otimes{\rm Pol}(F), we may write X_s as a finite sum X_s=\sum_kX_{s,k}\otimes y_k, where X_{s,k}\in M_{N_\kappa}(\mathbb{C})\otimes{\rm Pol}(H). Since x is reduced one has \sum_k E_\kappa(X_{s,k})y_k=(E_\kappa\otimes{\rm id})(X_s)=0. Since the family (y_k)_k is linearly independent we obtain E_\kappa(X_{s,k})=0 for all k. Write x as a finite sum:

where, in the last equality, we used the relations \rho_\kappa(f)=1\otimes1\otimes f, for all f\in C(F). Since E_F(a_sy_{k_s})=E_F(a_s)y_{k_s}=0 for all s, we deduce from the Claim that x\in\mathcal{C}_\kappa.

(4). By (1), it suffices to prove the formula for X\in\mathcal{C}_\kappa and for X\in M_{N_\kappa}(\mathbb{C})\otimes {\rm Pol}(H)\otimes {\rm Pol}(F). If X\in\mathcal{C}_\kappa we have ({\rm id}\otimes\Delta)(X)\in\mathcal{C}_\kappa\otimes C(\mathbb{G}) by (3). It now follows from (2) that

If now X\in M_{N_\kappa}(\mathbb{C})\otimes {\rm Pol}(H)\otimes {\rm Pol}(F) then, since \Delta\vert_{C(H)\otimes_{{\rm max}} C(F)}=\Delta_{H\times F},

and therefore (\mathbb{E}_\kappa\otimes{\rm id}_{C(\mathbb{G})})({\rm id}\otimes\Delta)(X) = ({\rm id}\otimes\Delta)(X) =({\rm id}\otimes\Delta)\mathbb{E}_\kappa.

(5). It suffices to apply ({\rm id}\otimes h_H\otimes h_F\otimes{\rm id}) to the equation proved in (4) and use the fact that the Haar state of H\times F is h_H\otimes h_F.

We start with the following Claim.

Claim. \ker(\widetilde{\mathbb{E}}_\kappa)=\overline{\mathcal{C}_\kappa+\ker(\psi_\kappa^{-1})\otimes {\rm Pol}(H)\otimes {\rm Pol}(F)}.

Proof of the Claim. Let x\in\ker(\widetilde{\mathbb{E}}_\kappa). By Lemma 4.3 (1), there are sequences (x_n) and (y_n) such that x_n\in \mathcal{C}_\kappa and y_n\in M_{N_\kappa}(\mathbb{C})\otimes {\rm Pol}(H)\otimes {\rm Pol}(F) such that x_n+y_n\rightarrow x. Hence,

Define y_n^\circ:=y_n-1_{M_{N_\kappa}(\mathbb{C})}\otimes(\psi_\kappa^{-1}\otimes{\rm id}\otimes{\rm id})(y_n) so that (\psi_\kappa^{-1}\otimes{\rm id}\otimes{\rm id})(y_n^\circ)=0. Note that y_n^\circ\in\ker(\psi_\kappa^{-1})\otimes{\rm Pol}(H)\otimes{\rm Pol}(F) and x_n+y_n^\circ\rightarrow x.\hfill \qed

End of the proof of Theorem. Suppose that x\in M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G}) then, since \widetilde{\Delta}(x)=({\rm id}\otimes\Delta)(x)\in M_{N_\kappa}(\mathbb{C})\otimes C_\kappa(\mathbb{G})\otimes C(\mathbb{G}) one has, by Lemma 4.3,

In particular, (E\otimes {\rm id})\widetilde{\Delta}(x)=\Delta_{H\times F}\circ E(x) for x\in C(H)\otimes_{{\rm max}} C(F)\subset C(\widetilde{\mathbb{G}}) and it remains to show that the formula holds for any reduced operator x in C(\widetilde{\mathbb{G}}). Write such a reduced operator x=x_1\dots x_n\in C(\widetilde{\mathbb{G}}), x_s\in M_{N_{\kappa_s}}(\mathbb{C})\otimes C_{\kappa_s}(\mathbb{G}) with \widetilde{\mathbb{E}}_{\kappa_s}(x_s)=0 and \kappa_s\neq\kappa_{s+1} so that E(x)=0. It suffices to show that (E\otimes {\rm id})\widetilde{\Delta}(x)=0. By linearity, continuity and the Claim, we may and will assume that, for all s, x_s\in\mathcal{C}_{\kappa_s}\cup\ker(\psi_\kappa^{-1})\otimes {\rm Pol}(H)\otimes{\rm Pol}(F). Since

\widetilde{\Delta}(x) is an alternating product from ({\rm id}\otimes\Delta)(\mathcal{C}_{\kappa_s})\subset\mathcal{C}_{\kappa_s}\otimes C(\mathbb{G})\subset \ker(\widetilde{\mathbb{E}}_{\kappa_s})\otimes C(\mathbb{G}) (Lemma 4.3) and we deduce from:

that (E\otimes{\rm id})\widetilde{\Delta}(x)=0. It proves the first statement of the Theorem.

(1). We use Remark 4.4 and the previous statement to deduce the first equation. Then, applying h_H\otimes h_F\otimes{\rm id} shows that (h_H\otimes h_F)\circ E\circ\iota is \Delta-invariant hence, h_\mathbb{G}=(h_H\otimes h_F)\circ E\circ\iota. We conclude the proof of (1) by Remark 2.6 (3).

(2). Let a\in C(G) and write \rho(a)=\sum_{\kappa,i,j} e^\kappa_{ij}\otimes\rho^\kappa_{ij}(a). We use the notations of Theorem 4.2, Lemma 4.3 and their proofs. By construction of E we have:

Hence, ({\rm id}_B\otimes E\circ\iota)\rho(a)=\sum_{\kappa,i}e^\kappa_{ii}\otimes 1_{C(H)}\otimes E_F(a)=1_B\otimes 1_{C(H)}\otimes E_F(a).

Consider the two canonical maps \beta_G\,:\, C(G)\otimes B\rightarrow B\otimes C(\mathbb{G}) and \beta_\Lambda\,:\, C(\Lambda)\otimes B\rightarrow B\otimes C(\mathbb{G}_\Lambda) and note that, by definition, ({\rm id}\otimes\iota_\Lambda)\beta_\Lambda=\beta_G\vert_{C(\Lambda)\otimes B}. It follows easily that

which implies that \iota_\Lambda intertwines the comultiplications. Let us now show that \iota_\Lambda is faithful. By Proposition 2.5, it suffices to show that \iota_\Lambda intertwines the Haar states (since it will imply that it is faithful at the reduced level). Note that, for 1\leq\kappa\leq K, \iota_\Lambda(C_\kappa(\mathbb{G}_\Lambda))\subset C_\kappa(\mathbb{G}). Hence, there exists a unique unital *-homomorphism \widetilde{\iota}_\Lambda\,:\, C(\widetilde{\mathbb{G}}_\Lambda)\rightarrow C(\widetilde{\mathbb{G}}) such that:

By construction, \widetilde{\iota}_\Lambda maps reduced operators in C(\widetilde{\mathbb{G}}_\Lambda) to reduced operators in C(\widetilde{\mathbb{G}}) and is the identity on C(H)\otimes_{{\rm max}} C(F). By uniqueness property, we have E\circ\iota_\Lambda=E_\Lambda, where E\,:\, C(\widetilde{\mathbb{G}})\rightarrow C(H)\otimes_{{\rm max}} C(F) and E_\Lambda\,:\, C(\widetilde{\mathbb{G}}_\Lambda)\rightarrow C(H)\otimes_{{\rm max}} C(F) are the canonical ucp maps constructed in Section ?. It follows from Theorem 4.5 that \iota_\Lambda intertwines the Haar states.

Write \mathbb{G}:=G\wr_{*,\beta,\widehat{\Gamma}} H with canonical morphism \rho\,:\, C(G)\rightarrow B\otimes C(\mathbb{G}) and view C(H),C^*(\Gamma)\subset C(\mathbb{G}). We also write \mathbb{G}':=\pi_1(\mathbb{G}_p,\mathbb{G}_e,\mathcal{G},\mathcal{T}), we view C(H)\subset C(\mathbb{G}_p)\subset C(\mathbb{G}') and we denote

View, for all p\in V(\mathcal{G}), \rho_p\,:\,C(G_p)\rightarrow B\otimes C(\mathbb{G}_p)\subset B\otimes C(\mathbb{G}') and note that, for all b\in C(\Lambda_e),

The universal property of C(G) gives a unique unital *-homomorphism

By construction of \rho', one has \rho'(C^*(\Gamma))\subset \mathbb{C} 1_B\otimes C(\mathbb{G}') hence, there exists a unique unital *-homomorphism \iota_{\widehat{\Gamma}}\,:\,C^*(\Gamma)\rightarrow C(\mathbb{G}') such that \rho'(f)=1_B\otimes\iota_{\widehat{\Gamma}}(f) for all f\in C^*(\Gamma). Note that for all p\in V(\mathcal{G}), a\in C(G_p) and b\in B, \rho'(a)\beta(b)=\rho_p(a)\beta(b)=\beta(b)\rho_p(a)=\beta(b)\rho'(a). Moreover, since each s_e' is the identity on C(H), each v_e commutes with C(H) inside C(\mathbb{G}') hence, \rho'(u_e)\beta(b)=(1_B\otimes v_e)\beta(b)=\beta(b)(1_B\otimes v_e) for all b\in B and e\in E(\mathcal{G}). Also, C(H) commutes with \iota_{\widehat{\Gamma}}(C^*(\Gamma)). By the universal property of the free wreath product C*-algebra C(\mathbb{G}), there exists a unique unital *-homomorphism \pi\,:\, C(\mathbb{G})\rightarrow C(\mathbb{G}') such that ({\rm id}\otimes\pi)\rho=\rho', \pi\vert_{C(H)} is the inclusion and \pi\vert_{C^*(\Gamma)}=\iota_{\widehat{\Gamma}}. Writing \beta_G\,:\, C(G)\otimes B\rightarrow B\otimes C(\mathbb{G}) and \beta_{G_p}\,:\, C(G_p)\otimes B\rightarrow B\otimes C(\mathbb{G}_p)\subset B\otimes C(\mathbb{G}') the canonical morphisms we have, by construction, \beta_{G_p}=({\rm id}\otimes\pi)\circ\beta_G\vert_{C(G_p)\otimes B}. For a\in C(G_p)\subset C(G) one has:

Since \Delta_{\mathbb{G}'}\pi(u_e)=\Delta_{\mathbb{G}'}(v_e)=v_e\otimes v_e=(\pi\otimes\pi)\Delta_{\mathbb{G}}(u_e) and,

we deduce that \pi intertwines the comultiplications.

We now show that \pi is an isomorphism by constructing its inverse morphism \pi^{-1}\,:\, C(\mathbb{G}')\rightarrow C(\mathbb{G}). We first note that, by the universal property of the free wreath product C*-algebra C(\mathbb{G}_p), there exists a unique \pi_p^{-1}\,:\, C(\mathbb{G}_p)\rightarrow C(\mathbb{G}) such that ({\rm id}\otimes\pi_p^{-1})\rho_p=\rho\vert_{C(\mathbb{G}_p)} and \pi^{-1}_p\vert_{C(H)} is the inclusion. Viewing C^*(\Gamma)\subset C(\mathbb{G}), we have the relation \rho(u_e)=1_B\otimes u_e, for all e\in E(\mathcal{G}). Note that, for all e\in E(\mathcal{G}) and b\in C(\Lambda_e) one has,

Moreover, \pi_{s(e)}^{-1}\circ s_e'\vert_{C(H)} is the inclusion of C(H)\subset C(\mathbb{G}) and since C(H) commutes with C^*(\Gamma) in C(\mathbb{G}), the universal property of C(\mathbb{G}') gives a unique unital *-homomorphism \pi^{-1}\,:\, C(\mathbb{G}')\rightarrow C(\mathbb{G}) such that \pi^{-1}\vert_{C(\mathbb{G}_p)}=\pi_p^{-1} and \pi^{-1}(v_e)=u_e. It is easy to check that \pi^{-1} is the inverse of \pi.

(1). By construction, \lambda_{H\times F}\circ E=E_r\circ\widetilde{\lambda}. Since E_r is faithful and \widetilde{\lambda} is surjective, we deduce that C_r(\widetilde{\mathbb{G}}) is the reduced C*-algebra of \lambda_{H\times F}\circ E and it follows from Remark 2.1 that \widetilde{\lambda}\circ\iota(C(\mathbb{G})) is the reduced C*-algebra of \lambda_{H\times F}\circ E\circ\iota=\lambda_{H\times F}\circ E_{H\times F}. We conclude the proof of (1) using Remark 2.6 (1).

(2). Since C(\mathbb{G}) is generated by the algebras C_\kappa(\mathbb{G}), for 1\leq\kappa\leq K, we conclude that C_r(\mathbb{G})=\widetilde{\lambda}\circ\iota(C(\mathbb{G})) is generated by C_{r,\kappa}(\mathbb{G})=\widetilde{\lambda}\circ\iota(C_\kappa(\mathbb{G})) (1\leq\kappa\leq K). Moreover, since the C_{r,\kappa}(\widetilde{\mathbb{G}}) are free with amalgamation with respect to E_r\vert_{C_{r,\kappa}(\widetilde{\mathbb{G}})}=(\psi_\kappa^{-1}\otimes{\rm id}\otimes{\rm id})\circ E_{2,\kappa} and since C_{r,\kappa}(\mathbb{G})\subset C_{r,\kappa}(\widetilde{\mathbb{G}}) the result follows.

(3). We construct the inverse isomorphism \varphi_{r,\kappa}\,:\, M_{N_\kappa(\mathbb{C})}\otimes C_{r,\kappa}(\mathbb{G})\rightarrow C_{r,\kappa}(\widetilde{\mathbb{G}}) satisfying \varphi_{r,\kappa}\circ({\rm id}\otimes(\widetilde{\lambda}\circ\iota)\vert_{C_\kappa(\mathbb{G})})=\widetilde{\lambda}_\kappa\circ\varphi_\kappa. We view both M_{N_\kappa}(\mathbb{C})\subset C_\kappa(\widetilde{\mathbb{G}}) and M_{N_\kappa}(\mathbb{C})\subset C_{r,\kappa}(\widetilde{\mathbb{G}}) in the obvious way. Note that, since \varphi_\kappa\circ\iota(C_\kappa(\mathbb{G}))=M_{N_\kappa}(\mathbb{C})'\cap C_\kappa(\widetilde{\mathbb{G}}) we deduce that C_{r,\kappa}(\mathbb{G})\subseteq M_{N_\kappa}(\mathbb{C})'\cap C_{r,\kappa}(\widetilde{\mathbb{G}}). Hence, there exists a unique faithful unital *-homomorphism \varphi_{r,\kappa}\,:\,M_{N_\kappa}(\mathbb{C})\otimes C_{r,\kappa}(\mathbb{G})\rightarrow C_{r,\kappa}(\widetilde{\mathbb{G}}), such that \varphi_{r,\kappa}(b\otimes x)=bx for all b\in M_{N_\kappa}(\mathbb{C}) and x\in C_{r,\kappa}(\mathbb{G}). Since C_\kappa(\widetilde{\mathbb{G}}) is generated by M_{N_\kappa}(\mathbb{C}) and \varphi_\kappa\circ\iota(C_\kappa(\mathbb{G})) and since \widetilde{\lambda}_\kappa is surjective we deduce that C_{r,\kappa}(\widetilde{\mathbb{G}}) is generated by M_{N_\kappa}(\mathbb{C}) and \widetilde{\lambda}_\kappa(\varphi_\kappa\circ\iota (C_\kappa(\mathbb{G})))=C_{r,\kappa}(\mathbb{G}). Both these C*-subalgebras are in the image of \varphi_{r,\kappa} so \varphi_{r,\kappa} is surjective.

(1)\Rightarrow(2). Suppose that C_r(G) and C_r(H) are exact then, since C_r(F)\subset C_r(G) is a C*-subalgebra, C_r(F) is also exact. Theorem 5.1 and [Dy04, Theorem 3.2] implies that C_r(\mathbb{G}) is exact.

(2)\Rightarrow(1). Assume that C_r(\mathbb{G}) is exact and fix 1\leq \kappa\leq K. Since C_{r,\kappa}(\mathbb{G})\subset C_r(\mathbb{G}) is a C*-subalgebra, we deduce that C_{r,\kappa}(\mathbb{G}) is exact so C_{r,\kappa}(\widetilde{\mathbb{G}})\simeq M_{N_\kappa}(\mathbb{C})\otimes C_{r,\kappa}(\mathbb{G}) is also exact. Finally, since both C_r(G), C_r(H)\subset C_{r,\kappa}(\widetilde{\mathbb{G}}) are C*-subalgebras, it concludes the proof.

(1). Let M:=\underset{{\rm L}^\infty(H)\otimes {\rm L}^\infty(F)}{\ast} (L_\kappa(\mathbb{G}),E''\vert_{L_\kappa(\mathbb{G})})\subset{\rm L}^\infty(\widetilde{\mathbb{G}}). By uniqueness of the reduced C*-algebra of E^r_{H\times F} we may and will view C_r(\mathbb{G})\subset M. Note that C_r(\mathbb{G})''=M and E''\vert_{C_r(\mathbb{G})}= E^r_{H\times F}. (1) follows.

(2). Can be proved as statement (3) in Theorem 5.1.

(1) and (2). The first equality follows from \psi\circ E_\kappa=\psi_\kappa^{-1}\otimes h_H (Proposition 2.13). The equality about modular groups follows from Remark 2.3 as well as statement (2).

(3). Fix a system of matrix units (e^\kappa_{ij})_{ij} in M_{N_\kappa}(\mathbb{C}) diagonalizing Q_\kappa i.e. Q_\kappa e^\kappa_{ij}=\lambda_{k,i}e^\kappa_{ij}. We note that L_\kappa(\mathbb{G})=M_{N_\kappa}(\mathbb{C})'\cap L_\kappa(\widetilde{\mathbb{G}}) is generated by {\rm L}^\infty(H)\otimes{\rm L}^\infty(F)\subset L_\kappa(\widetilde{\mathbb{G}}) and

\rho^\kappa_{ij}(a):=\sum_r(e^\kappa_{ri}\otimes 1\otimes 1)(1_B\otimes a)(e^\kappa_{jr}\otimes 1\otimes 1)\in L_\kappa(\widetilde{\mathbb{G}}). In particular, the linear span of

is a unital weakly dense *-subalgebra of L_\kappa(\mathbb{G}). Hence, it suffices to show that, for any X of the form X=x_0\rho^\kappa_{i_1j_1}(a_1)\dots\rho^\kappa_{i_nj_n}(a_n)x_n as above, one has \varphi_\kappa(e^\kappa_{st}X)=\psi_\kappa^{-1}(e^\kappa_{st})\varphi_\kappa(X), for all 1\leq s,t\leq N_\kappa. Note that e^\kappa_{st}X=(e^\kappa_{si_1}\otimes x_0)(1\otimes a_1)(e^\kappa_{j_1i_2}\otimes x_2)\dots(1\otimes a_n)(e^\kappa_{j_nt}\otimes x_n). Then,

where X':=(1\otimes a_1)(e^\kappa_{j_1i_2}\otimes x_2)\dots(1\otimes a_n) and

Note that \psi_\kappa^{-1}\left(\sum_re^\kappa_{ri_1}Ye^\kappa_{j_nr}\right)=(\psi_\kappa^{-1}\otimes h_H\otimes h_F)E''_{\kappa,2}(X)=\varphi_\kappa(X).

Writing e^\kappa_{j_ni_1}=\frac{\sum_r\lambda_{\kappa,r}^{-1}e^\kappa_{j_nr}e^\kappa_{ri_1}}{{\rm Tr}(Q_\kappa^{-1})} we get:

For the last part of (3) we note that, under the identification L_\kappa(\widetilde{\mathbb{G}})=M_{N_\kappa}(\mathbb{C})\otimes L_\kappa(\mathbb{G}), \sigma_t^{\varphi_\kappa}=\sigma_{-t}^{\psi_\kappa}\otimes\sigma_t^{h_\kappa} so, for all x\in L_\kappa(\mathbb{G}), \sigma_t^{h_{\widetilde{\mathbb{G}}}}(x)=\sigma_t^{\varphi_\kappa}(x)=(\sigma_{-t}^{\psi_\kappa}\otimes\sigma_t^{h_\kappa})(1\otimes x)=\sigma_t^{h_\kappa}(x).

(4). Since for all x\in L_\kappa(\mathbb{G}), (h_H\otimes h_F)\circ E''(x)=\varphi_\kappa(x)=h_\kappa(x), we deduce from general properties of the amalgamated free product that \sigma_t^{h_\mathbb{G}}(x)=\sigma_t^{h_\kappa}(x). Since {\rm L}^\infty(\mathbb{G}) is generated by the L_\kappa(\mathbb{G}), for 1\leq\kappa\leq K, it concludes the proof.

As in the proof of Proposition 5.4, we note that {\rm L}^\infty(\mathbb{G}) is generated by {\rm L}^\infty(H)\otimes{\rm L}^\infty(F) and \{\rho^\kappa_{ij}(a)\,:\,1\le\kappa\leq K,\,1\leq i,j\leq N_\kappa,\,a\in {\rm L}^\infty(G)\}. Since H\times F is a dual quantum subgroup of \mathbb{G} one has, by definition of \tau_t^{\widetilde{\mathbb{G}}}, \tau_t^{\widetilde{\mathbb{G}}}(x)=(\tau_t^H\otimes\tau_t^F)(x)=\tau_t^{\mathbb{G}}(x) for all x\in{\rm L}^\infty(H)\otimes{\rm L}^\infty(F). Moreover,

If follows that \tau_t^{\widetilde{\mathbb{G}}}({\rm L}^\infty(\mathbb{G}))={\rm L}^\infty(\mathbb{G}), for all t\in\mathbb{R}. A similar computation gives

It follows that ({\rm id}\otimes\tau_t^{\widetilde{\mathbb{G}}})\rho=(\sigma_{t}^\psi\otimes{\rm id})\rho\circ\tau_t^G and ({\rm id}\otimes\sigma_t^{h_\mathbb{G}})\rho=(\sigma_{t}^\psi\otimes{\rm id})\rho\circ\sigma_t^{h_G}.

Claim. For all t\in\mathbb{R} and all X\in {\rm Pol}(G)\otimes B one has \beta_G\circ(\tau_t^G\otimes\sigma_{-t}^\psi)(X)=(\sigma_{-t}^\psi\otimes\tau_t^{\widetilde{\mathbb{G}}})\circ\beta_G(X).

End of the proof of Proposition 5.6. To finish the proof, it suffices to show that, for all t\in\mathbb{R}, one has \Delta_{\mathbb{G}}\circ\sigma_t^{h_\mathbb{G}}(x)=(\tau_t^{\widetilde{\mathbb{G}}}\otimes\sigma_t^{h_\mathbb{G}})\circ\Delta_{\mathbb{G}}(x), for all x\in{\rm L}^\infty(H)\otimes{\rm L}^\infty(F) and all x\in L_\kappa(\mathbb{G}), 1\leq\kappa\leq K. If x\in{\rm L}^\infty(H)\otimes{\rm L}^\infty(F) the equality is clear, since H\times F is a dual quantum subgroup of \mathbb{G}. Moreover, using the claim we find, for all x\in{\rm Pol}(G),

It finishes the proof.

We first remark that the map restricted to {\rm Pol}(G) is faithful by the formula proved in Theorem 4.5,

and the fact that all the states are faithful. Moreover, on {\rm Pol}(G), it is defined as

therefore, the extension of \rho to {\rm L}^\infty(G) is also normal, and therefore the intertwining of the states still holds, and implies that \rho is still faithful at the von Neumann algebraic level.

Since H is a dual quantum subgroup of \mathbb{G}, we get T^\tau(\mathbb{G})\subset T^\tau(H) and, using the formulas \tau^H_t(u^{ij,\kappa}_{kk,\gamma}) = \left(\dfrac{\lambda_{\kappa,i}}{\lambda_{\kappa,j}} \right)^{it}u_{kk,\gamma}^{ij,\kappa} and \sigma_t^\psi(b^\kappa_{ij})=\left(\frac{\lambda_{\kappa,i}}{\lambda_{\kappa,j}}\right)^{it} we deduce that T^\tau(H)\subset B^\psi, where B^\psi denotes the centralizer of \psi. Fix t\in T^\tau(\mathbb{G}) then \sigma_t^\psi={\rm id} and,

From this we get that \tau_t^G={\rm id} by faithfulness of the map \rho. Therefore T^\tau(\mathbb{G}) \subset T^\tau(G)\cap T^\tau(H). Assume now that t \in T^\tau(G)\cap T^\tau(H), then t\in B^\psi and, using that ({\rm id}\otimes \tau^\mathbb{G}_t)\rho = (\sigma_t^\psi\otimes {\rm id})\rho\circ \tau_t^G = \rho, we get that \tau_t^\mathbb{G} is trivial on the coefficients of \rho. Since t\in T^\tau(H) we deduce that \tau_t^\mathbb{G}={\rm id}.

(1). Assume that {\rm L}^\infty(\mathbb{G}) has the Haagerup property. Since H is a dual quantum subgroup of \mathbb{G}, {\rm L}^\infty(H) is a subalgebra with expectation, by Lemma 2.5. Therefore {\rm L}^\infty(H) has the Haagerup approximation property. By the decomposition of {\rm L}^\infty(\mathbb{G}) as a free product, we get for any \kappa a faithful conditional expectation onto the algebra L_\kappa(\mathbb{G}), which implies that the algebra L_\kappa(\mathbb{G}) and therefore M_{N_\kappa}(\mathbb{C}) \otimes L_\kappa(\mathbb{G})=L_\kappa(\widetilde{\mathbb{G}}) have the Haagerup approximation property. Since there is a conditional expectation L_\kappa(\widetilde{\mathbb{G}}) \rightarrow (B\otimes {\rm L}^\infty(G)), B\otimes {\rm L}^\infty(G) has the Haagerup property, and therefore \widehat{G} has it too. For the converse, assume that {\rm Irr}(F) is finite and that \widehat{G} and \widehat{H} have the Haagerup property, then we can apply [CKSVW21, Corollary 8.2] to the decomposition of {\rm L}^\infty(\widetilde{\mathbb{G}}) as an inductive amalgamated free product over B\otimes {\rm L}^\infty(F) as described in Remark 5.9. First, we remark that by the result of [CKSVW21, Corollary 8.2], the algebra \ast_\kappa( M_{N_\kappa}(\mathbb{C}),\psi_\kappa^{-1}) has the Haagerup property and therefore, if \widehat{H} has the Haagerup approximation property, then so does M_0. Moreover if \widehat{G} has the Haagerup property, then so does B\otimes {\rm L}^\infty(G), and an induction shows that M_K \simeq {\rm L}^\infty(\widetilde{\mathbb{G}}) has the Haagerup property. Finally, using the conditional expectation {\rm L}^\infty(\widetilde{\mathbb{G}}) \rightarrow {\rm L}^\infty(\mathbb{G}), we get that {\rm L}^\infty(\mathbb{G}) has the Haagerup property.

(2). We already know thanks to Proposition 5.5 that \mathbb{G} is Kac if and only G and H are Kac. Under these condition, h_G, h_H,\psi,h_\mathbb{G} are all traces. Assume first that \widehat{\mathbb{G}} is hyperlinear then, as H is a dual quantum subgroup of \mathbb{G}, \widehat{H} is also hyperlinear. Note that B being finite dimensional (B,\psi) is Connes embeddable (CE). Hence, (B\otimes{\rm L}^\infty(\mathbb{G}),\psi\otimes h_\mathbb{G}) also is, by [Pop95]. Since the map \rho\,:\,{\rm L}^\infty(G)\rightarrow B\otimes{\rm L}^\infty(\mathbb{G}) is faithful and satisfies (\psi\otimes h_\mathbb{G})\rho=h_G, \widehat{G} is hyperlinear. Conversely, assume that both \widehat{G} and \widehat{H} are hyperlinear and {\rm L}^\infty(F) is amenable. Then, [BDJ08, Corollary 4.5] implies that *_\kappa(M_{N_\kappa}(\mathbb{C}),\psi_\kappa^{-1}) is CE hence, M_0 is CE. Applying [BDJ08, Corollary 4.5] (and since {\rm L}^\infty(F) is amenable) inductively in Remark 5.9, we deduce that ({\rm L}^\infty(\widetilde{\mathbb{G}}),h_{\widetilde{\mathbb{G}}}) is CE. Since the inclusion {\rm L}^\infty(\mathbb{G})\subset{\rm L}^\infty(\widetilde{\mathbb{G}}) is state preserving, we deduce that \widehat{\mathbb{G}} is hyperlinear.

Write {\rm Irr}(H)=\{1,x_1,x_2\}, where x_1\neq x_2 are non-trivial irreducibles. Since x_1 is non-trivial we have either \overline{x}_1=x_1 or \overline{x}_1=x_2. Assume that \overline{x}_1=x_1, so that \overline{x}_2=x_2. Hence, \mathrm{Mor}(1,x_1\otimes x_2)=0 and it follows that there exists n_1,n_2\geq 0 such that x_1\otimes x_2=n_1x_1+n_2x_2. Also, \dim\mathrm{Mor}(1,x_k\otimes x_k)=1, \dim\mathrm{Mor}(x_1,x_2\otimes x_2)=\dim\mathrm{Mor}(x_2,x_1\otimes x_2)=n_2 and \dim\mathrm{Mor}(x_2,x_1\otimes x_1)=\dim\mathrm{Mor}(x_1,x_1\otimes x_2)=n_1. Hence, there exists s,t\geq 0 such that x_1\otimes x_1=1+ sx_1+n_1x_2 and x_2\otimes x_2=1+n_2x_1+tx_2. It follows that d_1^2=1+sd_1+n_1d_2 and d_2^2=1+n_2d_1+td_2, where d_k:=\dim(x_k). In particular, {\rm gcd}(d_1,d_2)=1. Note that (n_1,n_2) is a solution to the Diophantine equation d_1d_2=n_1d_1+n_2d_2 and since {\rm gcd}(d_1,d_2)=1 and n_1,n_2\geq 0 we deduce that (n_1,n_2)\in\{(d_2,0),(0,d_1)\}. Assume that n_1=d_2 and n_2=0. Then, d_1^2=1+sd_1+d_2^2 and d_2^2=1+td_2. From the equation d_2^2-td_2-1=0 we deduce that \Delta_2:=t^2+4 is a perfect square, which happens if and only if t=0. We conclude that d_2=1. From the equation d_1^2-sd_1-2=0 we deduce that \Delta_1:=s^2+8 is a perfect square, which happens if and only if s=1. Hence, d_1=2. It follows that \dim(C(H))=1+d_1^2+d_2^2=6. Assuming now that n_1=0 and n_2=d_1, the same proof, exchanging the role of x_1 and x_2 also shows that \dim(C(H))=6. However, the only compact quantum groups whose associated C*-algebras have dimension 6 are \mathbb{Z}_6, S_3 \simeq \mathbb{Z}_3\rtimes\mathbb{Z}_2 and its dual \widehat{S_3}. \mathbb{Z}_6 and \widehat{S_3} both have 6 irreducibles, each of dimension 1, and therefore H \simeq S_3. It remains to check the other case, that is \overline{x}_1=x_2 and \overline{x}_2=x_1. Let s,t\geq 0 be such that x_2\otimes x_1=1+tx_1+sx_2 and write d=\dim(x_1)=\dim(x_2). Hence, d^2=1+(t+s)d and, this implies that 1=0 modulo d so d=1 and x_2\otimes x_1=1. Letting u\in\mathcal{U}(C(H)) the group like unitary u:=x_1\neq 1 we have u^*=x_2\neq u so u^3=1 and we deduce that H=\mathbb{Z}_3.

(1)\Rightarrow(2). View \beta\,:\,\mathbb{Z}_p\rightarrow{\rm Aut}(B)\subset\mathcal{L}(B) and write \mathbb{Z}_p=\langle\sigma\rangle where \sigma^p=1. Hence \beta(\sigma)^p={\rm id}_B so {\rm Sp}(\beta(\sigma))\subseteq\{\xi^k\,:\, 0\leq k\leq p-1\}, where \xi=e^{2i\pi/p} and we may write B=\bigoplus_{k=0}^{p-1}B_k as vector spaces, where B_k=\{b\in B:\beta(\sigma)b=\xi^kb\}. Since \beta is ergodic we have B_0=\mathbb{C} 1_B. Note that B_k^*=B_{-k} and B_kB_l\subset B_{k+l}, where the sum is modulo p. If B_{1}\neq\{0\}, take a nonzero a\in B_{1}. Then, a^*a\in B_0=\mathbb{C}1 and is non-zero. Let \lambda\neq 0 be such that a^*a=\lambda1. Take \lambda^{-1/2} be any square root of \lambda^{-1} and define u:=\lambda^{-1/2}a. Then u satisfies u^*u=1_B and, since B is finite dimensional, we also have uu^*=1. Note that, for any 1\leq k\leq p-1, any b\in B_{k}, one has b(u^{k})^*\in B_kB_{-k}\subset B_0=\mathbb{C}1_B which implies that B_{k}=\mathbb{C} u^{k}. We also have u^p\in (B_1)^p\subset B_0=\mathbb{C}1. Hence, up to replacing u by a modulus one multiple of u, we may and will assume that u^p=1. Hence, the unique map C^*(\mathbb{Z}_p)\rightarrow B=\bigoplus_{k=0}^{p-1}\mathbb{C} u^k which map \sigma onto u is an isomorphism intertwining the left translation on C^*(\mathbb{Z}_p) with the action \beta. Assume now that B_1=\{0\}. Fix 2\leq k\leq p-1 and assume that B_k\neq 0. Let a\in B_k be non-zero and, as before, we may further assume by ergodicity that a^*a=aa^*=1. Then, by Fermat’s Little Theorem, a^{k^{p-1}-1}\in B_k^{k^{p-1}-1}\subset B_{k^{p-1}}=B_1=\{0\}. Hence, a^{p^{k-1}-1}=0 and a^*=(a^*)^{p^{k-1}}a^{p^{k-1}-1}=0 so a=0 a contradiction. We deduce that B_k=0 for all 1\leq k\leq p-1. So B=\mathbb{C}1 and \beta is the trivial action.

(2)\Rightarrow(1). if p>2 is not prime write p=nm with n,m\geq 2 and consider the ergodic action \mathbb{Z}_n\curvearrowright \mathbb{C}^n=C^*(\mathbb{Z}_n) by left translation. Then, composing with the canonical surjection \mathbb{Z}_p\rightarrow\mathbb{Z}_n sending the generator of \mathbb{Z}_p on the generator of \mathbb{Z}_n, we get an ergodic action \mathbb{Z}_p\curvearrowright\mathbb{C}^n. However, this action is not faithful so it can not be isomorphic to the left translation of \mathbb{Z}_p, which is faithful.

Write B=B_1\oplus B_\sigma\oplus B_{\sigma^2}\oplus B_\tau\oplus B_{\tau\sigma}\oplus B_{\sigma\tau}, where B_g:=\{b\in B\,:\,\beta(b)=b\otimes g\}. The order 2 elements are \{\tau,\tau\sigma,\sigma\tau\}. Note that A_\sigma:=B_1\oplus B_\sigma\oplus B_{\sigma^2} is a subalgebra of B and \beta(A_\sigma)\subset A_\sigma\otimes C^*(\sigma) so we get an ergodic action \mathbb{Z}_3\curvearrowright A_\sigma. Similarly, for any order two element \omega\in\{\sigma,\tau\sigma,\sigma\tau\}, A_\omega:=B_1\oplus B_\omega is a subalgebra with an ergodic action \mathbb{Z}_2\curvearrowright A_\omega.

Assume first that \mathbb{Z}_3\curvearrowright A_\sigma is ergodic then there is an element u\in B_\sigma such that u^3=1 and B_\sigma=\mathbb{C} u, B_{\sigma^2}=\mathbb{C} u^2. Now, if for some \omega\in\{\tau,\sigma\tau,\tau\sigma\} the action \mathbb{Z}_2\curvearrowright A_\omega is also ergodic then we have B_\tau=\mathbb{C} v with some v satisfying v^2=1. Since B_gB_h\subset B_{gh} and \langle\sigma,\omega\rangle=S_3 we deduce that B_g\neq\{0\} for all g\in S_3. It follows that \dim(B)=6 and \beta is isomorphic to \beta_4. If now \mathbb{Z}_2\curvearrowright A_\omega is trivial for all \omega\in \{\tau,\tau\sigma,\sigma\tau\}. It follows that \dim(B)=3 and \beta is isomorphic to \beta_3.

Assume now that \mathbb{Z}_3\curvearrowright A_\sigma is the trivial action. Since for any \omega_1,\omega_2\in\{\tau,\tau\sigma,\sigma\tau\} with \omega_1\neq\omega_2 we have \langle\omega_1,\omega_2\rangle=S_3 and since B_\sigma=\{0\}, there at most one \omega\in\{\tau,\sigma\tau,\tau\sigma\} such that \mathbb{Z}^3\curvearrowright B_\omega is nontrivial. If there is such a \omega then \beta is isomorphic to \beta_2 and if there is no such \omega then \beta is the trivial action \beta_1 on B=\mathbb{C}.

(1)\Rightarrow(2) follows from the general theory [To06].

(2)\Rightarrow(3). Since {\rm L}^\infty(\mathbb{G}) is amenable, each L_\kappa(\widetilde{\mathbb{G}})=M_{N_\kappa}(\mathbb{C})\otimes L_\kappa(\mathbb{G}) is amenable. Write C=B\otimes{\rm L}^\infty(G) and D=M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H) so that L_\kappa(\widetilde{\mathbb{G}})=C\underset{B}{*}D. Let p:=e^\gamma_{jj}\in B be a minimal projection and note that \sigma_t^\psi(p)=p for all t\in\mathbb{R}. Let us show that pCp={\rm L}^\infty(G) and pDp=\beta_\kappa(p)(M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H))\beta_\kappa(p) are *-free in pL_\kappa(\widetilde{\mathbb{G}})p with respect to the state \omega:=(\varphi_\kappa(p))^{-1}\varphi_\kappa(p\cdot p). Let x\in pL_\kappa(\widetilde{\mathbb{G}}) p be a word alternating from pCp and pDp of the form x=(pc_1p)(pd_1p)\dots (pc_np)(pd_np), with c_k\in C, d_k\in D and \omega(pc_kp)=\omega(pd_kp)=0. Since p\in B is a minimal projection, we have for e_k\in\{c_k,d_k\}, E_B(pe_k p)=pE_B(e_k)p\in\mathbb{C}1_B. Hence, E_B(pe_k p)=\psi(E_B(pe_kp))1_B=\varphi_\kappa(pe_kp)1_B=0. It follows that x is a reduced operator in the amalgamated free product L_\kappa(\widetilde{\mathbb{G}})=C\underset{B}{*}D. Hence, \varphi_\kappa(x)=0=\psi(p)\omega(x) so \omega(x)=0. One proves similarly that \omega(x)=0 for other alternating products x. This shows that pCp*pDp\simeq\langle pCp,pDp\rangle\subset p L_\kappa(\widetilde{\mathbb{G}}) p, where the free product is with respect to the state \omega. Note that, since p\in B^\psi\subset (L_\kappa(\widetilde{\mathbb{G}}))^{\varphi_\kappa}, one has \sigma_t^\omega(pxp)=p\sigma_t^{\varphi_\kappa}(x)p, for all t\in\mathbb{R}, x\in L_\kappa(\widetilde{\mathbb{G}}). In particular, \sigma^\omega_t(pCp)=pCp and \sigma^\omega_t(pDp)=pDp. Hence, \sigma_t^\omega(\langle pCp,pDp\rangle)=\langle pCp,pDp\rangle and, by Takesaki’s Theorem, the subalgebra \langle pCp,pDp\rangle\subset p L_\kappa(\widetilde{\mathbb{G}})p has a faithful normal conditional expectation. Hence, pCp*pDp={\rm L}^\infty(G)*\beta_\kappa(p)(M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H))\beta_\kappa(p) is amenable. Note that if {\rm L}^\infty(H) is infinite dimensional, then {\rm L}^\infty(H) is diffuse by [FT24, Lemma 2.4]. Hence, M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H) is also diffuse and \beta_\kappa(p)(M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H))\beta_\kappa(p) must be infinite dimensional. Since {\rm L}^\infty(G)\neq\mathbb{C}, the results of [Ue10] imply that pCp*pDp is not amenable. Hence, {\rm L}^\infty(H) is finite dimensional. Let q_x be the minimal central projections of {\rm L}^\infty(H), which are parameterized by x\in{\rm Irr}(\widehat{H}) and note that \varepsilon_H(a)q_1=aq_1 for all a\in C(H), where 1\in{\rm Irr}(\widehat{H}) is the trivial representation. Note that

Indeed, if the dimension is 1, then \beta_\kappa(p) is a minimal projection and it implies that \beta_\kappa(p)(1_\kappa\otimes p_1)=\beta_\kappa(p). Hence,

Since \beta_\kappa is faithful, we deduce that K=1 and \beta(p)=p\otimes p_1. Applying \psi\otimes{\rm id} we get \psi(p)1=\psi(p)p_1 so p_1=1 and H is trivial, a contradiction.

Since pCp,pDp\neq\mathbb{C} and pCp*pDp is amenable, the results of [Ue10] imply that:

We deduce that \dim({\rm L}^\infty(G))= 2=\dim(\beta_\kappa(p)(M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H))\beta_\kappa(p)) which implies that G\simeq\mathbb{Z}_2. Hence both G and H are Kac, by [VD97, Theorem 2.2]. Hence, \mathbb{G} is Kac and, since {\rm L}^\infty(\mathbb{G}) is amenable and \mathbb{G} is Kac, [To06] implies that \mathbb{G} is amenable so G^{*K}, as a compact quantum subgroup of \mathbb{G}, is also amenable. [FT24, Proposition 2.8] implies that K\leq 2. Write

Write, for 1\leq\kappa\leq K, I_\kappa:=\{x\in{\rm Irr}(\widehat{H}):\beta_\kappa(p)(1\otimes q_x)\neq 0\} and note that, since (q_x) are central and pairwise orthogonal, \{\beta_\kappa(p)(1\otimes q_x):x\in I_\kappa\} is a family of non-zero pairwise orthogonal projections in \beta_\kappa(p)(M_{N_\kappa}(\mathbb{C})\otimes {\rm L}^\infty(H))\beta_\kappa(p). Hence,

Note that {\rm Irr}(\widehat{H})= \cup_{\kappa=1}^K I_\kappa. Indeed, writing I:=\{x\in{\rm Irr}(\widehat{H}):\beta(p)(1\otimes p_x)\neq 0\} one has:

Since \psi(p)>0 it follows that \sum_{x\in I}p_x=1 so I={\rm Irr}(\widehat{H}). Let x\in{\rm Irr}(\widehat{H}), since \beta(p)(1\otimes q_x)\neq 0, there exists \kappa such that x\in I_\kappa and this shows that {\rm Irr}(\widehat{H})= \cup_{\kappa=1}^K I_\kappa.

If K=1 this shows that \vert{\rm Irr}(\widehat{H})\vert=2. Arguing as in the proof of [FT24, Proposition 2.8], we deduce that \widehat{H}\simeq \mathbb{Z}_2. Hence, H\simeq\widehat{\mathbb{Z}_2}\simeq\mathbb{Z}_2.

If K=2. Since ({\rm id}\otimes\varepsilon_H)\beta={\rm id}_B, the trivial representation 1 is in I_1\cap I_2. Hence, \vert {\rm Irr}(\widehat{H})\vert=\vert I_1\vert+\vert I_2\vert-\vert I_1\cap I_2\vert\leq \vert I_1\vert+\vert I_2\vert -1\leq 3 so either \vert {\rm Irr}(\widehat{H})\vert=2 in which case we deduce again that H=\mathbb{Z}_2 or \vert {\rm Irr}(\widehat{H})\vert=3 and Lemma 5.11 implies that \widehat{H}\in\{\mathbb{Z}_3,S_3\} so H\in\{\mathbb{Z}_3,\widehat{S}_3\}. In conclusion, we have G=\mathbb{Z}_2 and either K=1 and H=\mathbb{Z}_2 or K=2 and H\in\{\mathbb{Z}_2,\mathbb{Z}_3,\widehat{S_3}\}. However, in the case K=2 and H=\mathbb{Z}_3, there is no ergodic action of \mathbb{Z}_3 on B, by Lemma 5.12. If K=2 and H=\widehat{S_3} then, by Lemma 5.13 we have B=\mathbb{C}^2 and \beta is the action \mathbb{C}^2\rightarrow\mathbb{C}^2\otimes C(\mathbb{Z}_2)=\mathbb{C}^2\otimes C^*(\omega)\subset\mathbb{C}^2\otimes C^*(S_3) coming from the unique ergodic action of \mathbb{Z}_2 on \mathbb{C}^2 and an order two element \omega\in S_3. We deduce from Proposition 3.8 that \mathbb{G}\simeq (\mathbb{Z}_2\wr_{*}\mathbb{Z}_2)\underset{\mathbb{Z}_2}{*}\widehat{S_3}. By Proposition 3.11 and [FT24, Proposition 2.23] we have \mathbb{Z}_2\wr_{*}\mathbb{Z}_2\simeq \widehat{\mathbb{Z}_2^{*2}}\rtimes \mathbb{Z}_2. Hence, {\rm L}^\infty(\mathbb{G})=({\rm L}(\mathbb{Z}_2^{*2})\otimes{\rm L}(\mathbb{Z}_2))\underset{{\rm L}(\mathbb{Z}_2)}{*}{\rm L}(S_3), where the amalgamated free product is with respect to the conditional expectations on the dual quantum subgroup \mathbb{Z}_2. Write S_3=\langle\sigma,\tau\vert\sigma^3=1,\tau^2=1,\tau\sigma\tau=\sigma^{-1}\rangle and let p:=\frac{1}{2}(1+\tau)\in{\rm L}(\mathbb{Z}_2) and note that p=p^*=p^2 and p\tau=p so p is a minimal projection in {\rm L}(\mathbb{Z}_2). However, p is not minimal in {\rm L}(S_3) since p{\rm L}(S_3)p=\mathbb{C} p\oplus\mathbb{C} p\sigma p. It is easy to check that {\rm L}(\mathbb{Z}_2^{*2})=p({\rm L}(\mathbb{Z}_2^{*2})\otimes{\rm L}(\mathbb{Z}_2))p and p{\rm L}(S_3)p are *-free in p{\rm L}^\infty(\mathbb{G})p with respect to h_\mathbb{G}(p)^{-1}h_\mathbb{G}\vert_{p{\rm L}^\infty(\mathbb{G})p}. Hence, {\rm L}(\mathbb{Z}_2^{*2})*p{\rm L}(S_3)p is isomorphic to a von Neumann algebra of p{\rm L}^\infty(\mathbb{G})p which is impossible since {\rm L}(\mathbb{Z}_2^{*2}) is diffuse and \dim(p{\rm L}(S_3)p)\geq 2 so {\rm L}(\mathbb{Z}_2^{*2})*p{\rm L}(S_3)p is not amenable. It follows that the only possible case is G=H=\mathbb{Z}_2 and Lemma 5.12 implies that B\in\{\mathbb{C},\mathbb{C}^2\}.

(3)\Rightarrow(1). If G=H=\mathbb{Z}_2 and B=\mathbb{C} then \beta is the trivial action on \mathbb{C} and it follows from Remark 3.7 that \mathbb{G}\simeq\widehat{\mathbb{Z}_2^{*2}} which is coamenable. If B=\mathbb{C}^2 then Lemma 5.12 implies that \beta is the unique ergodic action on B=\mathbb{C}^2 so that it is isomorphic to the universal action \mathbb{Z}_2=S_2^+\curvearrowright\mathbb{C}^2. Proposition 3.11 and [FT24, Proposition 2.23] imply that \mathbb{G}\simeq\widehat{\mathbb{Z}_2^{*2}}\rtimes \mathbb{Z}_2, which is coamenable by [FT24, Proposition 2.17].

Let \varphi\in N_* be a faithful state and fix an orthonormal basis (e_i)_i of N for the \varphi scalar product. Then, E(a) = \sum_{i} \varphi(e_i^* E(a))e_i=\sum_{i} \varphi\circ E(e_i^* a)e_i. For all x,y\in M,

and this goes to 0, because u_n goes to 0 weakly and the sum is finite. The same holds for the conjugate, because a\mapsto a^* is weakly continuous. It concludes the proof.

Let n\geq 1, a projection q\in M_n(\mathbb{C})\otimes M_1, a non-zero partial isometry w\in (M_{1,n}(\mathbb{C})\otimes M)q and a unital normal *-homomorphism \pi\,:\, Q\rightarrow q(M_n(\mathbb{C})\otimes M_1)q such that \pi(Q)\subset q(M_n(\mathbb{C})\otimes M_1)q is with expectation and xw=w\pi(x) for all x\in Q.

Claim: One has w^*w\in qM_n( M_1)q and w^*\mathcal{N}_M(Q)''w\subset w^*wM_n(M_1)w^*w.

Proof of the Claim. We work in the amalgamated free product:

with respect to the conditional expectation \widetilde{E}_k:={\rm id}_{M_{n}(\mathbb{C})}\otimes E_k, k=1,2. Since \pi(Q) is diffuse we may apply [HU16b, Lemma 2.1] to get a f.n.s. \varphi on \pi(Q) with diffuse centralizer \pi(Q)^\varphi and note that q=\pi(1)\in\pi(Q)^\varphi. Define the normal state \psi_1:=\varphi\circ E(q\,\cdot\,q)\in (M_n(\mathbb{C})\otimes M_1)_*, where E\,:\, q (M_n(\mathbb{C})\otimes M_1)q\rightarrow Q is any faithful normal conditional expectation. Fix any f.n.s. \omega\in [(1-q)M(1-q)]_* and define the f.n.s. \psi_2:=\frac{1}{2}\psi_1(\cdot)+\frac{1}{2}\omega((1-q)\cdot(1-q))\in (M_n(\mathbb{C})\otimes M_1)_*. Then, the f.n.s. \psi:=\psi_2\circ({\rm id}\otimes E_{M_1})\in (M_n(\mathbb{C})\otimes M)_* satisfies the following: \psi=\psi\circ({\rm id}\otimes E_{M_1}), q\in (M_n(\mathbb{C})\otimes M_1)^\psi and \pi(Q)\cap q(M_n(\mathbb{C})\otimes M_1)^\psi q is diffuse. By Lemma 6.6 there is a sequence of unitaries (u_k)_k\in \pi(Q)\cap q(M_n(\mathbb{C})\otimes M_1)^\psi q such that \widetilde{E}_1(xu_ky)\rightarrow 0 in norm, for all x,y\in q(M_n(\mathbb{C})\otimes M_1)q. Viewing the sequence (u_k)_k in the unit ball of (M_n(\mathbb{C})\otimes M_1)^\psi we can apply Lemma 6.8 and deduce that ({\rm id}\otimes E_{M_1})(xu_ky)\rightarrow 0 \sigma-strongly, for all x,y\in q(M_n(M)\ominus M_n(M_1))q. Fix a normalizing unitary x\in\mathcal{N}_M(Q) and observe that, for all a\in Q, one has w^*xw\pi(a)=w^*xaw=\pi(xax^*)w^*xw. Define the element v:=w^*xw-({\rm id}\otimes E_{M_1})(w^*xw)\in qM_n(M)q and note that ({\rm id}\otimes E_{M_1})(v)=0. Moreover, for all a\in Q, one has v\pi(a)=\pi(xax^*)v. For any k\in\mathbb{N}, fix a unitary a_k\in Q such that \pi(a_k)=u_k. Then, for all k\in\mathbb{N}, since \pi(xa_kx^*) is a unitary in \pi(Q),

It follows that vv^*=0 so w^*xw\in M_n(M_1) for all x\in \mathcal{N}_M(Q). This proves the Claim.

End of the proof of Lemma 6.9. By Remark 6.3 (2), it suffices to show that

Since xw=w\pi(x) for all x\in Q, we deduce that ww^*\in Q'\cap M\subset\mathcal{N}_M(Q)''. Define the projections e:=e_{11}\otimes ww^*\in M_n(\mathbb{C})\otimes\mathcal{N}_M(Q)'', f:=w^*w\in M_n(\mathbb{C})\otimes M_1 and note that, since w is a partial isometry we have, viewing w\in (e_{11}\otimes 1)(M_n(\mathbb{C})\otimes M),

so w\in e(M_n(\mathbb{C})\otimes M)f. By the Claim, we have a unital normal *-homomorphism

defined by \theta(e_{11}\otimes x)=w^*xw for all x\in ww^*\mathcal{N}_M(Q)''ww^*. Moreover, w\theta(x)=ww^*xw=xw for all x\in ww^*\mathcal{N}_M(Q)''ww^*. Finally, note that the image of \theta is exactly w^*\mathcal{N}_M(Q)''w and, given any faithful normal conditional expectation E\,:\, M\rightarrow\mathcal{N}_M(Q)'', the map

is obviously a normal faithful conditional expectation.

Since Q\subset M_1 is unital with expectation and Q\nprec_{M_1}N we deduce from Remark 6.3 (2) that M_1\nprec_{M_1} N so that we can directly apply [HI20, Theorem 5.2].

Assume that P:=\mathcal{N}_M(A)''\nprec_MM_k for k=1,2 and let us show that P is amenable. By Lemma 6.9 we have A\nprec M_k for k=1,2. Hence, we may apply [HU16a, Theorem 4.4] and deduce that A'\cap M is amenable. Let \mathcal{A}:=A\vee(A'\cap M)\subset M and note that \mathcal{A} is amenable and diffuse. Also, by Remark 6.1, \mathcal{A}\subset M is with expectation. Define \mathcal{P}:=\mathcal{N}_M(\mathcal{A})'' and note that P\subset\mathcal{P}. By Remark 6.1 P\subset M is with expectation so P\subset\mathcal{P} is with expectation and it suffices to show that \mathcal{P} is amenable. Note that \mathcal{A}'\cap M\subset A\subset \mathcal{A} and since \mathcal{A} is diffuse and N is finite dimensional, \mathcal{A}\nprec_M N. Also, since the inclusion P\subset\mathcal{P} is unital with expectation, Remark 6.3 (2) implies that \mathcal{P}\nprec_M M_k for k=1,2. Note that the statement of [IS22, Theorem G] is still valid, using normalizers instead of stable normalizers (and the proof is even easier, by using [HU16a, Theorem A.4] instead of [IS22, Lemma 6.1]). Hence, we deduce that \mathcal{P} is amenable.

We write M=M_i\underset{N}{*}M_i'. Let us show that the inclusion N\subset M_i' is entirely non trivial. If there is a non-zero projection z\in\mathcal{Z}(N) such that M_i'z=Nz then M_jz\subset M_i'z=Nz which implies that M_jz=Nz. Since N\subset M_j is entirely non-trivial, we deduce that N\subset M_i' also is. Hence, we may apply [HI20, Theorem A] to deduce (1), (2) and (4). It remains to show (3) for which we can repeat the proof of [Ue13, Theorem 4.3 (3)]: fix any faithful normal state \omega\in N_* and consider the state \varphi:=\omega\circ E_N. Then, M is semifinite if and only if there exists one-parameter group of unitaries u_t\in M such that \sigma_t^\varphi={\rm Ad}(u_t) for all t\in \mathbb{R}. Fix i such that M_i\nprec_{M_i} N. Applying Lemma 6.10 we deduce that u_t\in M_i, for all t\in\mathbb{R}. Let now j\neq i be such that N\subset M_j is entirely non-trivial. Let y\in M_j such that E_j(y)=0 and note that \sigma_t^\varphi(y)(u_t-E_i(u_t))+ \sigma_t^\varphi(y)E_i(u_t)=\sigma_t^\varphi(y)u_t=u_ty=E_i(u_t)y+(u_t-E_i(u_t))y. Writing v:=u_t-E_i(u_t) we have, by freeness of M_i and M_j and since \sigma_t^\varphi(y)\in\ker(E_j) and v\in\ker(E_i),

It follows that E_N(vyy^*v^*)=E_N(vE_j(yy^*)v^*)=0 and since E_N is faithful, vE_j(yy^*)=0, for all y\in\ker(E_j). Hence, vs(E_j(yy^*))=0, for all y\in\ker(E_j), where s(z) denotes the support projection of the self-adjoint operator z. By [Ue13, Lemma 4.2 (3)], there is a family (y_i)_{i\in I} of elements in \ker(E_j) such that \sum_{i\in i}s(E_j(y_iy_i^*))=1. It implies that v=0 i.e. u_t=E_i(u_t)\in N. Write u_t=h^{it} with h a positive non-singular operator affiliated with the centralizer N^\omega\subset N of \sigma_t^\omega. Then, the semifinite faithful normal weight \omega_{h^{-1}} on N is the desired trace.

Since {\rm Irr}(G) is infinite, it follows from [FT24, Lemma 2.4] that {\rm L}^\infty(G) is diffuse. Hence, {\rm L}^\infty(G)q is also diffuse. Moreover, since {\rm L}^\infty(F)q\subset{\rm L}^\infty(G)q is a finite dimensional subalgebra and h_G':=h_G(q)^{-1}h_G\vert_{{\rm L}^\infty(G)q} is an almost periodic state on {\rm L}^\infty(G)q (note that q is in the centralizer of h_G), Remark 6.7 implies that there exists a sequence of unitaries (u_n)_n in ({\rm L}^\infty(G)q)^{h'_G} such that E_F(xu_ny)\rightarrow 0 in norm, for all x,y\in{\rm L}^\infty(G)q. Considering the amalgamated free product:

Lemma 6.8 (2) implies that E_{M_2}(x(1\otimes u_n)y)\rightarrow 0 \sigma-strong for all x,y\in M. We conclude from Remark 6.5 that M\nprec_{M}M_2. Remark 6.3 (3) implies that

(1). It suffices to prove the statement for L_\kappa(\widetilde{\mathbb{G}})=M_{N_\kappa}(\mathbb{C})\otimes L_{\kappa}(\mathbb{G}) instead of L_\kappa(\mathbb{G}). Since {\rm Irr}(G) and {\rm Irr}(H) are infinite, it follows from [FT24, Lemma 2.4] that both {\rm L}^\infty(G) and {\rm L}^\infty(H) are diffuse. Hence, since B\otimes{\rm L}^\infty(F) is finite dimensional,

and, B\otimes{\rm L}^\infty(F)\subset B\otimes{\rm L}^\infty(G) is entirely non-trivial.

From the discussion above, Theorem 6.12 (1) and Lemma 2.20 (2) we find that

Let u\in\mathcal{Z}(L_\kappa(\widetilde{\mathbb{G}})) and write u=1\otimes v, with v\in\mathcal{Z}({\rm L}^\infty(F)). Then, for all x\in{\rm L}^\infty(G), u(1_B\otimes x)u^*=1_B\otimes vxv^*=1_B\otimes x, which implies that v\in\mathcal{Z}({\rm L}^\infty(G))\otimes{\rm L}^\infty(F). The converse inclusion being trivial, this gives the formula for the center. Note that L_\kappa(\widetilde{\mathbb{G}}) has no amenable direct summand by Theorem 6.12 (4).

(2). Let p\in\mathcal{Z}(L_\kappa(\mathbb{G})) be a non-zero projection. Write p=1\otimes q with q\in\mathcal{Z}({\rm L}^\infty(G))\cap{\rm L}^\infty(F). Note that:

Since {\rm L}^\infty(G)q and {\rm L}^\infty(H) are diffuse and B\otimes{\rm L}^\infty(F)q is finite dimensional, we can apply Theorem 6.12 (2) and we find:

Fix t\in T(L_\kappa(\widetilde{\mathbb{G}})(1\otimes p)) and let v\in\mathcal{U}(B\otimes{\rm L}^\infty(F)q) be such that \sigma_t^{\varphi_\kappa}\vert_{L_\kappa(\widetilde{\mathbb{G}})(1\otimes p)}={\rm Ad}(v). Note that, since F is a finite quantum group, the Haar state of F is tracial by [VD97, Theorem 2.2] so \sigma_t^{h_F}={\rm id} hence, restricting the modular group of \varphi_\kappa to B\otimes{\rm L}^\infty(H)\otimes{\rm L}^\infty(F)q, we find that \sigma^{\psi_\kappa}_{-t}\otimes\sigma_t^{h_H}\otimes{\rm id}_{{\rm L}^\infty(F)q}={\rm Ad}((\beta_\kappa\otimes{\rm id})(v)). Lemma 2.20 implies that v=1\otimes u for some unitary u\in\mathcal{Z}({\rm L}^\infty(F)q) and \sigma_t^{h_H}={\rm id}. In particular, \sigma_t^\psi={\rm id} and, restricting this time the modular group of \varphi_\kappa to B\otimes{\rm L}^\infty(G)q gives \sigma_t^\psi\otimes\sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}={\rm id}\otimes\sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}={\rm Ad}(1\otimes u). Hence, T(L_\kappa(\widetilde{\mathbb{G}})(1\otimes p))\subseteq T', where:

Let now t\in T'. Then \sigma_t^{h_H}={\rm id} implies that \sigma_t^\psi={\rm id}_B and \sigma_{-t}^{\psi_\kappa}={\rm id}_{M_{N_\kappa}(\mathbb{C})}. Let u\in\mathcal{U}(\mathcal{Z}({\rm L}^\infty(F)q)) be such that \sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}={\rm Ad}(u). Then,

since u\in\mathcal{U}(\mathcal{Z}({\rm L}^\infty(F)q)). It follows that t\in T(L_\kappa(\widetilde{\mathbb{G}})(1\otimes p)).

We write {\rm L}^\infty(\mathbb{G})=\underset{{\rm L}^\infty(H)\otimes{\rm L}^\infty(F),1\leq\gamma\leq K}{*}L_\gamma(\mathbb{G}) and since L_\gamma(\mathbb{G})\nprec_{L_\gamma(\mathbb{G})}{\rm L}^\infty(H)\otimes{\rm L}^\infty(F) (Lemma 6.13) we have in particular that the inclusion {\rm L}^\infty(H)\otimes{\rm L}^\infty(F)\subset L_\gamma(\mathbb{G}) is entirely non-trivial, for 1\leq\gamma\leq K.

(1). Note that we always have \mathbb{C}1_{{\rm L}^\infty(H)}\otimes ({\rm L}^\infty(F)\cap\mathcal{Z}({\rm L}^\infty(G)))\subseteq\mathcal{Z}({\rm L}^\infty(\mathbb{G})). If K=1 statement (1) follows directly from Lemma 6.14 (since {\rm dim}(B)\geq 3). Hence we may and will assume that K\geq 2 so that we can apply Theorem 6.12 to deduce that {\rm L}^\infty(\mathbb{G}) has no amenable direct summand. Let u\in\mathcal{U}(\mathcal{Z}({\rm L}^\infty(\mathbb{G}))). If there exists 1\leq\kappa\leq K such that N_\kappa\geq 2 then, since u\in \mathcal{N}_{{\rm L}^\infty(\mathbb{G})}(L_\kappa(\mathbb{G})), it follows from Lemma 6.10 that u is in the center of L_\kappa(\mathbb{G}). We deduce that u\in\mathbb{C}1_{{\rm L}^\infty(H)}\otimes ({\rm L}^\infty(F)\cap\mathcal{Z}({\rm L}^\infty(G))) by Lemma 6.14. Hence, we may and will assume that N_\kappa=1 for all \kappa. Hence, B=\mathbb{C}^K and K\geq 3.

Let u\in\mathcal{U}(\mathcal{Z}({\rm L}^\infty(\mathbb{G}))). By Theorem 6.12, u\in{\rm L}^\infty(\mathbb{G})'\cap{\rm L}^\infty(H)\otimes{\rm L}^\infty(F). In particular, u\in\mathcal{Z}(L_1(\mathbb{G}))\cap\mathcal{Z}(L_2(\mathbb{G})). For each 1\leq\kappa\leq K, we can again apply Theorem 6.12 to the amalgamated free product L_\kappa(\mathbb{G})=(\mathbb{C}^K\otimes{\rm L}^\infty(G))\underset{\mathbb{C}^K\otimes{\rm L}^\infty(F)}{*}({\rm L}^\infty(H)\otimes{\rm L}^\infty(F)) to deduce:

In particular, u\in(\beta_1(\mathbb{C}^K)\otimes{\rm L}^\infty(F))\cap(\beta_2(\mathbb{C}^K)\otimes{\rm L}^\infty(F))=\mathbb{C}1\otimes{\rm L}^\infty(F), by Lemma 2.21. Then, there exists v\in{\rm L}^\infty(F) such that u=1\otimes v and since u is central in \mathbb{C}^K\otimes{\rm L}^\infty(G), we have v\in\mathcal{Z}({\rm L}^\infty(G)). Hence, \mathcal{Z}({\rm L}^\infty(\mathbb{G}))\subseteq \mathbb{C}1\otimes\mathcal{Z}({\rm L}^\infty(G))\cap{\rm L}^\infty(F).

(2). Once again, we may and will assume that K\geq 2, By Lemma 6.14. Let p=1\otimes q be a projection in \mathcal{Z}({\rm L}^\infty(\mathbb{G})), with q\in\mathcal{Z}({\rm L}^\infty(G))\cap{\rm L}^\infty(F). Note that p is in the centralizer of h_{\mathbb{G}} and q is in the centralizer of h_G and:

Since L_\kappa(\mathbb{G})p\nprec_{L_\kappa(\mathbb{G})p}{\rm L}^\infty(H)\otimes{\rm L}^\infty(F)q (Lemma 6.13) we have \mathcal{N}_{{\rm L}^\infty(\mathbb{G})p}( L_\gamma(\mathbb{G})p)\subset L_\gamma(\mathbb{G})p for all 1\leq\gamma\leq K, by Lemma 6.10. Define:

Let t\in T' and let u\in\mathcal{U}(\mathcal{Z}({\rm L}^\infty(F)q)) such that \sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}={\rm Ad}(u). Since \sigma_t^{h_H}={\rm id} one has \sigma_t^\psi={\rm id} and \sigma_{-t}^{\psi_\gamma}={\rm id} for all 1\leq\gamma\leq K. Arguing as in the end of the proof of Lemma 6.14, we deduce that \sigma_t^{\varphi_\gamma}\vert_{L(\widetilde{\mathbb{G}}_\gamma)(1\otimes p)}={\rm Ad}(v), where

Hence \sigma_t^{\varphi_\gamma}\vert_{L(\widetilde{\mathbb{G}}_\gamma)(1\otimes p)} is actually implemented by the same unitary in {\rm L}^\infty(F)q, for all 1\leq\gamma\leq K. Since the isomorphism M_{N_\gamma}(\mathbb{C})\otimes L_\gamma(\mathbb{G})p\simeq L_\gamma(\widetilde{\mathbb{G}})(1\otimes p) intertwines \psi_\gamma^{-1}\otimes h_\gamma\vert_{L_\gamma(\mathbb{G})p}, it intertwines the modular automorphisms \sigma_t^{\psi_\gamma^{-1}}\otimes\sigma_t^{h_\gamma}\vert_{L_\gamma(\mathbb{G})p}={\rm id}_{M_{N_\gamma}(\mathbb{C})}\otimes\sigma_t^{h_\gamma}\vert_{L_\gamma(\mathbb{G})p} and \sigma_t^{\varphi_\gamma}\vert_{L_\gamma(\widetilde{\mathbb{G}})(1\otimes p)} and since the isomorphism is the inclusion on {\rm L}^\infty(H)\otimes{\rm L}^\infty(F)q, {\rm id}_{M_{N_\gamma}(\mathbb{C})}\otimes\sigma_t^{h_\gamma}\vert_{L_\gamma(\mathbb{G})p} and hence \sigma_t^{h_\gamma}\vert_{L_\gamma(\mathbb{G})p} is implemented by the same unitary in {\rm L}^\infty(F)q, for all 1\leq\gamma\leq K. Since \sigma_t^{h_\mathbb{G}}\vert_{{\rm L}^\infty(\mathbb{G})p}=\underset{1\leq \gamma\leq K}{*}\sigma_t^{h_\gamma}\vert_{L_\gamma(\mathbb{G})p}, we deduce that t\in T({\rm L}^\infty(\mathbb{G})p).

Let us prove the converse inclusion. Let t\in T({\rm L}^\infty(\mathbb{G})p). Then, there exists v\in\mathcal{U}({\rm L}^\infty(\mathbb{G})p) such that \sigma_t^{h_\mathbb{G}}\vert_{{\rm L}^\infty(\mathbb{G})p}={\rm Ad}(v). Since \sigma_t^{h_\mathbb{G}}({L_\gamma(\mathbb{G})})=\sigma_t^{h_\gamma}(L_\gamma(\mathbb{G}))=L_\gamma(\mathbb{G}), we have v\in\mathcal{N}_{{\rm L}^\infty(\mathbb{G})p}(L_\gamma(\mathbb{G})p)\subset L_\gamma(\mathbb{G})p, for all 1\leq\gamma\leq K. If there exists 1\leq\kappa\leq K such that N_\kappa\geq 2 then it follows that t\in T(L_\kappa(\mathbb{G})p) so t\in T', by Lemma 6.14. Hence, we may and will assume that N_\kappa=1, for all 1\leq\kappa\leq K. In that case, B=\mathbb{C}^K, K\geq 3 and \psi is the uniform trace since it is the unique \delta-form on \mathbb{C}^K. Viewing, for all 1\leq\gamma\leq K,

and observing that:

we may apply Lemma 6.10 to deduce that v\in (\mathbb{C}^K\otimes{\rm L}^\infty(G)q)\cap({\rm L}^\infty(H)\otimes{\rm L}^\infty(F)q)\subset L_\gamma(\mathbb{G})q, for all 1\leq \gamma\leq K. In particular, v\in\beta_1(\mathbb{C}^K)\otimes{\rm L}^\infty(F)q\cap\beta_2(\mathbb{C}^K)\otimes{\rm L}^\infty(F)q=\mathbb{C}1\otimes{\rm L}^\infty(F)q, by Lemma 2.21. Write v=1\otimes u, where u\in\mathcal{U}({\rm L}^\infty(F)q). Since, \sigma_t^\psi\otimes\sigma_t^{h_G}={\rm id}\otimes\sigma_t^{h_G}={\rm Ad}(1\otimes u) we deduce that \sigma_t^{h_G}={\rm Ad}(u) and u\in\mathcal{Z}({\rm L}^\infty(F)) (since F is Kac). We also have \sigma_t^{h_H}\otimes\sigma_t^{h_F}={\rm Ad}((\beta_1\otimes{\rm id})(1\otimes u))={\rm id}, which implies that \sigma_t^{h_H}={\rm id}.

(3). We fix a projection p\in\mathcal{Z}({\rm L}^\infty(\mathbb{G})) and we write p=1\otimes q, where q\in \mathcal{Z}({\rm L}^\infty(G))\cap{\rm L}^\infty(F) is a projection.

(a)\Rightarrow(b). Assume that H is Kac and there exists Q\in\mathcal{Z}({\rm L}^\infty(F)q) positive and invertible such that \sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}={\rm Ad}(Q^{it}). Then \psi and \psi_\kappa^{-1} are traces so

since Q\in\mathcal{Z}({\rm L}^\infty(F)q). Also, \sigma_t^{\varphi_\kappa}\vert_{B\otimes{\rm L}^\infty(G)q}={\rm id}\otimes\sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}={\rm Ad}(1\otimes Q^{it}). Hence, \sigma_t^{\varphi_\kappa}\vert_{L_\kappa(\mathbb{G})p}={\rm Ad}(Q_1^{it}), where Q_1:=(1\otimes Q)=(\beta_\kappa\otimes{\rm id})(1\otimes Q)\in L_\kappa(\widetilde{\mathbb{G}})p, for all 1\leq\kappa\leq K. Arguing as in the end of the proof of (2), we deduce that \sigma_t^{h_\kappa}\vert_{L_\kappa(\mathbb{G})p} is implemented by the same unitary Q_1^{it}\in{\rm L}^\infty(F)q, for all 1\leq\kappa\leq K and so \sigma_t^{h_\mathbb{G}}\vert_{{\rm L}^\infty(\mathbb{G})p} is also implemented by Q_1^{it}\in{\rm L}^\infty(F)q, for all t\in\mathbb{R}. It follows that the faithful state \tau=h_{\mathbb{G}}(p)^{-1}h_\mathbb{G}(Q_1^{-1}\cdot)\vert_{{\rm L}^\infty(\mathbb{G})p} is a trace on {\rm L}^\infty(\mathbb{G})p.

(b)\Rightarrow(c) is obvious.

(c)\Rightarrow(a). Assume that {\rm L}^\infty(\mathbb{G})p is semifinite. Hence, T({\rm L}^\infty(\mathbb{G})p)=\mathbb{R} and it follows from (2) that H is Kac. In particular, \psi is a tracial \delta-form so it is the uniform trace on B i.e. \psi=\frac{1}{K}\sum_\kappa\frac{1}{N_\kappa}{\rm Tr}_\kappa (and \psi_\gamma is tracial, for all 1\leq\gamma\leq K). Moreover, since L_\gamma(\mathbb{G})p\subset{\rm L}^\infty(\mathbb{G})p has expectation, L_\gamma(\mathbb{G})p is semifinite, for all 1\leq\gamma\leq K, so L_\gamma(\widetilde{\mathbb{G}})(1\otimes p)=M_{N_\gamma}(\mathbb{C})\otimes L_\gamma(\mathbb{G})p also is. Write L_\gamma(\widetilde{\mathbb{G}})(1\otimes p)=(B\otimes{\rm L}^\infty(G)q)\underset{B\otimes{\rm L}^\infty(F)q}{*}(M_{N_\gamma}(\mathbb{C})\otimes{\rm L}^\infty(H)\otimes{\rm L}^\infty(F)q) and apply Theorem 6.12 (3) to deduce that there exists a faithful normal semifinite trace \omega on B\otimes{\rm L}^\infty(F)q such that \omega\circ({\rm id}\otimes E_F) is tracial on B\otimes{\rm L}^\infty(G)q. Note that, since B\otimes{\rm L}^\infty(F)q is finite dimensional, \omega is actually finite and up to renormalization, we may and will assume that \omega is a faithful normal tracial state on B\otimes{\rm L}^\infty(F)q. Then, there exists Q_0\in\mathcal{Z}(B\otimes{\rm L}^\infty(F)q) positive and invertible such that \omega=(\psi\otimes h_{F})(Q_0\cdot). Write Q_0=\sum_{\gamma,r}e^\gamma_{rr}\otimes Q_\gamma, where Q_\gamma\in\mathcal{Z}({\rm L}^\infty(F)q) is positive and invertible. In particular, Q:=\frac{1}{K}\sum_\gamma Q_\gamma\in\mathcal{Z}({\rm L}^\infty(F)q) is positive and invertible and, for all x\in {\rm L}^\infty(G)q, one has:

Note that the modular group of h_G(Q\cdot)\vert_{{\rm L}^\infty(G)q} is given by Q^{it}\sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}(\cdot)Q^{-it}. Since \omega\circ({\rm id}\otimes E_F) is tracial, the faithful state h_G(Q\cdot)\vert_{{\rm L}^\infty(G)q} also is hence Q^{it}\sigma_t^{h_G}(\cdot)\vert_{{\rm L}^\infty(G)q}Q^{-it}={\rm id} and \sigma_t^{h_G}\vert_{{\rm L}^\infty(G)q}={\rm Ad}(Q^{-it}), for all t\in\mathbb{R}.

Let t\in T^\tau_{\rm Int}(\mathbb{G}) and u_0\in \mathcal{U}({\rm L}^\infty(\mathbb{G})) such that \tau^\mathbb{G}_{t} = {\rm Ad}(u_0). Since \tau^\mathbb{G}_t preserves the subalgebras L_\kappa(\mathbb{G}), and \mathcal{N}_{{\rm L}^\infty(\mathbb{G})}(L_\kappa(\mathbb{G}))''\subset L_\kappa(\mathbb{G}) we deduce that u_0\in L_\kappa(\mathbb{G}), for all 1\leq\kappa\leq K.

First assume that, for some 1\le\kappa\leq K, N_\kappa\geq 2. Identifying L_\kappa(\widetilde{\mathbb{G}}) with M_{N_\kappa}(\mathbb{C})\otimes L_\kappa(\mathbb{G}) and \widetilde{\tau}^\kappa_t:=\tau_t^{\widetilde{\mathbb{G}}}\vert_{L_\kappa(\widetilde{\mathbb{G}})} with \sigma_{-t}^{\psi_\kappa}\otimes\tau_t^{\mathbb{G}}\vert_{L_\kappa(\mathbb{G})} we can write \widetilde{\tau}^\kappa_t={\rm Ad}(w), where w=v\otimes u_0 for some v\in\mathcal{U}(M_{N_\kappa}(\mathbb{C})). Viewing {\rm L}^\infty(\widetilde{\mathbb{G}}) as an amalgamated free product and since, by construction, \widetilde{\tau}^\kappa_t leaves globally invariant both subalgebras B\otimes{\rm L}^\infty(G) and M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(H)\otimes{\rm L}^\infty(F) we deduce that there exists u in the amalgam M_{N_\kappa}(\mathbb{C})\otimes{\rm L}^\infty(F) such that \widetilde{\tau}^\kappa_t={\rm Ad}(u). In particular, \sigma_{-t}^{\psi_\kappa}\otimes\tau_t^H\otimes{\rm id}={\rm Ad}((\beta_\kappa\otimes{\rm id})(u)) (since \tau_t^F is trivial, because {\rm L}^\infty(F) is finite dimensional). Lemma 2.20 gives that u\in \mathbb{C}1\otimes \mathcal{Z}({\rm L}^\infty(F)) and \tau_t^H={\rm id}. Hence, viewing u\in \mathcal{Z}({\rm L}^\infty(F)) we have \tau_t^G={\rm Ad}(u).

Now assume that N_\kappa=1, for all 1\leq\kappa\leq K so that B=\mathbb{C}^K, {\rm L}^\infty(\widetilde{\mathbb{G}})={\rm L}^\infty(\mathbb{G}), K\geq 3 and \psi is the uniform trace since it is the unique \delta-form on \mathbb{C}^K. Viewing, for all 1\leq\gamma\leq K,

and since \tau_t^\mathbb{G} leaves globally invariant both \mathbb{C}^K\otimes{\rm L}^\infty(G) and {\rm L}^\infty(H)\otimes{\rm L}^\infty(F), we may apply Lemma 6.10 to deduce that u_0\in (\mathbb{C}^K\otimes{\rm L}^\infty(G))\cap({\rm L}^\infty(H)\otimes{\rm L}^\infty(F))\subset L_\gamma(\mathbb{G}), for all 1\leq \gamma\leq K. In particular, u_0\in\beta_1(\mathbb{C}^K)\otimes{\rm L}^\infty(F)\cap\beta_2(\mathbb{C}^K)\otimes{\rm L}^\infty(F)=\mathbb{C}1\otimes{\rm L}^\infty(F), by Lemma 2.21. Write u_0=1\otimes u, where u\in\mathcal{U}({\rm L}^\infty(F)). Since

we deduce that \tau_t^{h_G}={\rm Ad}(u) and u\in\mathcal{Z}({\rm L}^\infty(F)) (since {\rm L}^\infty(F) is finite dimensional). We also have:

which implies that \tau_t^{H}={\rm id}. This shows the first inclusion.

Let now t\in\mathbb{R} be such that \tau_t^H={\rm id} and u\in\mathcal{U}(\mathcal{Z}({\rm L}^\infty(F))) such that \tau_t^G={\rm Ad}(u). By construction of \tau_t^{\mathbb{G}}, we can argue as the in proof of the T-invariant computation in Theorem 6.15 to deduce that \tau_t^{\mathbb{G}}={\rm Ad}(u).

The fact that {\rm L}^\infty(\mathbb{G}) is a non-amenable factor of type {\rm II}_1 or {\rm III} follows from Theorem 6.15. Using the notations of Remark 5.9, we write {\rm L}^\infty(\widetilde{\mathbb{G}}) as an amalgamated free product:

To simplify the notations we write M for M_{K-1}. We then have the following Claim.

Claim. One has {\rm L}^\infty(\mathbb{G})\nprec_{{\rm L}^\infty(\widetilde{\mathbb{G}})} M and {\rm L}^\infty(\mathbb{G})\nprec_{{\rm L}^\infty(\widetilde{\mathbb{G}})} \nu_K(B\otimes{\rm L}^\infty(G)).

Proof of the Claim. Since {\rm Irr}(G) is infinite, the von Neumann algebra {\rm L}^\infty(G) is diffuse [FT24, Lemma 2.4] and since the Haar state of G is almost periodic, its centralizer {\rm L}^\infty(G)^{h_G} is still diffuse. Let (u_n)_n be a sequence of unitaries in {\rm L}^\infty(G)^{h_G} going to 0 \sigma-weakly. Since {\rm L}^\infty(\widetilde{\mathbb{G}})=\nu_K(B\otimes{\rm L}^\infty(G))*_{B\otimes{\rm L}^\infty(F)}M we may apply Lemma 6.6 to deduce that

Since \nu_K(1\otimes u_n) is in the centralizer of h_{\widetilde{\mathbb{G}}} for all n, we may apply Lemma 6.8 to deduce that E_{M}(x\nu_K(1\otimes u_n)y)\rightarrow 0 \sigma-strongly for all x,y\in {\rm L}^\infty(\widetilde{\mathbb{G}}). In particular, E_{M}(x\rho^K_{ij}(u_n)y)\rightarrow 0 \sigma-strongly for all x,y\in {\rm L}^\infty(\widetilde{\mathbb{G}}) and 1\leq i,j\leq N_\kappa. Note that the map

is a unital *-homomorphism so that \rho_K(u_n) is unitary in M_{N_K}(\mathbb{C})\otimes{\rm L}^\infty(\mathbb{G}). Moreover,

for all X,Y\in M_{N_\kappa}(\mathbb{C})\otimes {\rm L}^\infty(\widetilde{\mathbb{G}}). Since M_{N_K}(\mathbb{C})\otimes{\rm L}^\infty(\mathbb{G}) is finite or of type {\rm III} we have

Hence, {\rm L}^\infty(\mathbb{G})\nprec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}M.

Now, since {\rm Irr}(H) is infinite and arguing as in the beginning of the proof, we have a sequence of unitaries (v_n)_n in {\rm L}^\infty(H)^{h_H} going to 0 \sigma-weakly so 1\otimes v_n\otimes 1\in M_{N_K}(\mathbb{C})\otimes{\rm L}^\infty(H)\otimes{\rm L}^\infty(F) is going to 0 \sigma-weakly and, by Lemma 6.6, we deduce that (E_K\otimes{\rm id}_F)(x(1\otimes v_n\otimes 1)y)\rightarrow 0 in norm, for all x,y\in M_{N_K}(\mathbb{C})\otimes{\rm L}^\infty(H)\otimes{\rm L}^\infty(F). Next, since \mu(v_n\otimes 1)\in M is in the centralizer of h_{\widetilde{\mathbb{G}}} for all n, we can apply Lemma 6.8 and deduce that E_{\nu_K(B\otimes{\rm L}^\infty(G))}(x\mu(v_n\otimes 1)y)\rightarrow 0 \sigma-strongly for all x,y\in{\rm L}^\infty(\widetilde{\mathbb{G}}). Note that \mu(v_n\otimes 1) is actually a sequence of unitaries in {\rm L}^\infty(\mathbb{G}). Since {\rm L}^\infty(\mathbb{G}) is finite or type {\rm III}, we deduce that {\rm L}^\infty(\mathbb{G})\nprec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}\nu_K(B\otimes{\rm L}^\infty(G)).\qed

End of the proof of Theorem 6.20.(1). By contradiction, assume that {\rm L}^\infty(\mathbb{G})=P\otimes Q, where P and Q are two diffuse factors. Since {\rm L}^\infty(\mathbb{G}) is either finite or type {\rm III} we may and will assume that both P and Q are either finite or of type {\rm III}. Moreover, since {\rm L}^\infty(\mathbb{G}) is not amenable one of them, let say P, is not amenable. Note that, since {\rm L}^\infty(\mathbb{G})=P\otimes Q, there is a normal faithful conditional expectation from {\rm L}^\infty(\mathbb{G}) to Q (and to P) and since {\rm L}^\infty(\mathbb{G})\subset{\rm L}^\infty(\widetilde{\mathbb{G}}) has expectation, we deduce that both Q, P\subset{\rm L}^\infty(\widetilde{\mathbb{G}}) have expectation. Since Q is diffuse, we may apply [HU16b, Lemma 2.1] to get a f.n.s. \omega\in Q_* with diffuse centralizer A:=Q^\omega\subset Q. Note that A\subset Q has expectation by Takesaki’s Theorem hence A\subset{\rm L}^\infty(\widetilde{\mathbb{G}}) also has expectation. Also, P\subset Q'\cap{\rm L}^\infty(\mathbb{G})\subset A'\cap{\rm L}^\infty(\mathbb{G})\subset A'\cap{\rm L}^\infty(\widetilde{\mathbb{G}}). Since P is not amenable and has expectation, A'\cap{\rm L}^\infty(\widetilde{\mathbb{G}}) is not amenable either. Since A is finite, diffuse and has expectation we may apply [HU16a, Theorem 4.4] to deduce that either A\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}\nu_K(B\otimes{\rm L}^\infty(G)) or A\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}M. Applying now Lemma 6.9 we deduce that \mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(A)''\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}\nu_K(B\otimes{\rm L}^\infty(G)) or \mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(A)''\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}M. Since P\subset A'\cap{\rm L}^\infty(\widetilde{\mathbb{G}}), Remark 6.3 (4) implies that P\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}\nu_K(B\otimes{\rm L}^\infty(G)) or P\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}M. Applying again Lemma 6.9 we get that \mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(P)''\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}\nu_K(B\otimes{\rm L}^\infty(G)) or \mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(P)''\prec_{{\rm L}^\infty(\widetilde{\mathbb{G}})}M. However, P,Q\subset\mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(P)'' which implies that {\rm L}^\infty(\mathbb{G})\subset\mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(P)'' so that, with Remark 6.3 (4) and the Claim, we obtain a contradiction.

(2). By contradiction, let A\subset{\rm L}^\infty(\mathbb{G}) be a Cartan subalgebra (with expectation). Since {\rm L}^\infty(\mathbb{G})\subset{\rm L}^\infty(\widetilde{\mathbb{G}}) has expectation, A\subset{\rm L}^\infty(\widetilde{\mathbb{G}}) has expectation. Note that {\rm L}^\infty(\mathbb{G})=\mathcal{N}_{{\rm L}^\infty(\mathbb{G})}(A)''\subset \mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(A)''. Hence, the Claim implies that

Moreover, since {\rm L}^\infty(\mathbb{G}) is not amenable and has expectation, \mathcal{N}_{{\rm L}^\infty(\widetilde{\mathbb{G}})}(A)'' is not amenable either. We get a contradiction with Lemma 6.11.

We use the counit, which is a *-morphism \varepsilon : C(G)\rightarrow \mathbb{C}, which gives \varepsilon_\ast : K_0(C(G)) \rightarrow K_0(\mathbb{C})\simeq \mathbb{Z}, sending [1] to [1], which is the generator of the K-theory group of \mathbb{C}. Therefore the class cannot be of torsion. If G is K-amenable, the same holds for the reduced C*-algebra, because \lambda_\ast, which is an isomorphism in that case, sends the class [1] to [1] too.

The Kasparov product of m_A and \gamma can be taken as [(\mathbb{C}^N \otimes A)\otimes_A H, {\rm id} \otimes_\eta 1, \Tilde{F}], where \Tilde{F} is an adequate F-connection making it a Kasparov triple. There is a unitary isomorphism

Under this isomorphism, the map {\rm id}\otimes_\eta 1 corresponds to the map U({\rm id}\otimes_\eta 1)U^* which sends any element b\otimes a\in M_N(\mathbb{C})\otimes A to b\otimes \eta(a)\in \mathcal{L}(\mathbb{C}^N\otimes H). We then fix \Tilde{F} to be the operator U (1\otimes F) U^*, and we only have to check that \Tilde{F} is a F-connection and that it defines a Kasparov module. First, for any b\otimes a\in M_N(\mathbb{C})\otimes A, we have

which is a compact operator because [\eta(a), F] is, by hypothesis, and \mathbb{C}^N is finite dimensional. The same way, (1\otimes F)^2 ({\rm id}\otimes \eta)(b\otimes a) = b\otimes F^2\eta(a) and

are also compact operators. Lastly, we have to check the connection properties. For any x\in (\mathbb{C}^N \otimes A), we define the adjointable map T_x \in \mathcal{L}(H, (\mathbb{C}^N \otimes A)\otimes_A H), \xi\mapsto x\otimes_\eta \xi. Then if x= f\otimes a, the map UT_x becomes \xi \mapsto f \otimes \eta(a)\xi hence:

For any \xi, we have (T_x F - \Tilde{F}T_x)\xi = U^*(f\otimes (\eta(a)F\xi-F\eta(a)\xi)), which is a compact operator by hypothesis. The other property can be shown the same way. Thus this indeed defines a Kasparov product for m_A\otimes \gamma, which corresponds to [\mathbb{C}^N\otimes H, {\rm id}\otimes \eta, 1\otimes F].

(1)\Rightarrow (2). Assume first that G and H are K-amenable, then we can apply [FG18, Theorem 5.1] to the amalgamated free product

to get that the full algebra C_\kappa(\widetilde{\mathbb{G}}) is K-equivalent to the reduced C_{r,\kappa}(\widetilde{\mathbb{G}}) for all \kappa, through the canonical map \widetilde{\lambda_\kappa} and as C_{\bullet,\kappa}(\widetilde{\mathbb{G}})\simeq M_{N_\kappa}(\mathbb{C})\otimes C_{\bullet,\kappa}(\mathbb{G}), the same is true for the map \lambda_\kappa such that \widetilde{\lambda_\kappa}={\rm id} \otimes \lambda_\kappa. We then apply it to

to get that \mathbb{G} is K-amenable.

(2)\Rightarrow (1). Assume that \mathbb{G} is K-amenable i.e. there exists \gamma \in KK(C_r(\mathbb{G}),\mathbb{C}) such that [\lambda_\mathbb{G}] \otimes \gamma = [\varepsilon_\mathbb{G}]\in KK(C(\mathbb{G}),\mathbb{C}). Then, since H is a dual quantum subgroup of \mathbb{G}, it is also K-amenable. We then fix any 1\leq \kappa\leq K, and we consider the morphism \pi :C(G)\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C(\mathbb{G}) defined by \pi : a \mapsto \beta_\kappa(e^\kappa_{11})\rho_\kappa(a). Note that ({\rm id}\otimes \varepsilon_\mathbb{G})\circ \pi = \varepsilon_G(\cdot)e_{11}^\kappa. By Theorem 4.5 we have ({\rm id}_B\otimes h_\mathbb{G})\rho=h_G(\cdot)1_B. In particular, ({\rm id}\otimes h_\mathbb{G})\rho_\kappa=h_G(\cdot)1. It follows that there exists a unique unital *-homomorphism \rho_{r,\kappa}\,:\,C_r(G)\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C_r(\mathbb{G}) such that ({\rm id}\otimes\lambda_\mathbb{G})\rho_\kappa=\rho_{r,\kappa}\lambda_G. Viewing \beta_\kappa(e^\kappa_{11}) as a projection in (M_{N_\kappa}(\mathbb{C})\otimes{\rm Pol}(H))\cap\rho_{r,\kappa}(C_r(G))', the map \rho_{r,\kappa}(C_r(G))\rightarrow M_{N_\kappa}(\mathbb{C})\otimes C_r(\mathbb{G}), x\mapsto\beta_\kappa(e^\kappa_{11})x is a (non-unital) *-homomorphism. Hence \pi_r: C_r(G) \rightarrow M_{N_\kappa}(\mathbb{C})\otimes C_r(\mathbb{G}) defined by \pi_r(a):=\beta_\kappa(e^\kappa_{11})\rho_{r,\kappa}(a) is such that \pi_r\circ \lambda_G = ({\rm id} \otimes \lambda_\mathbb{G})\circ \pi. Let us show that \Tilde{\gamma} := [\pi_r] \otimes m_{C_r(\mathbb{G})} \otimes \gamma \in KK(C_r(G),\mathbb{C}) satisfies [\lambda_G]\otimes \Tilde{\gamma} =[\varepsilon_G]. Using Lemma 7.2 we find:

because the action of e_{11}^\kappa is only on the first coordinate in \mathbb{C}^{N}, and the second summand is a degenerate module. Hence [\lambda_G]\otimes \Tilde{\gamma} =[\varepsilon_G] and G is K-amenable.

We write

and we can immediately apply the result of [FG20] to get the exact sequences for the reduced and the full case. Now assume that {\rm Irr}(F) is finite. Then C(F)\simeq C_r(F)\simeq {\rm Pol}(F) is a finite dimensional C*-algebra, therefore we have immediately K_1(B\otimes C_\bullet(F))=0 and so the vertical map on the right-hand side is 0. We can also deduce from this that

and we also have in any case

Using that \iota_\ast : K_0(C_\bullet(F))\rightarrow K_0(C_\bullet(G)) is injective, we see that the first horizontal map must be injective in the full case. This also implies that the left vertical map is 0.

The exact sequence is a direct application of [FG20] to the amalgamated free product. The map K_1(C_\bullet(H\times F)) \rightarrow K_1(C_{\bullet,\kappa}(\mathbb{G})) corresponds to \iota_\ast, while the map in the part of the isomorphism K_1(M_{N_\kappa}(\mathbb{C})\otimes C_\bullet(H\times F)) \hookrightarrow K_1(M_{N_\kappa}(\mathbb{C})\otimes C_{\bullet,\kappa}(\mathbb{G})) corresponds to ({\rm id}\otimes \iota)_\ast, which is injective as shown in Proposition 7.4. These maps are the same at the level of the K-theory, through the Morita equivalence. Therefore the map at the utmost right on the bottom line is injective, and the map on the right is 0. Assume then that F is trivial. We have a sequence of morphisms

where \pi_H is defined in Remark 3.7. The composition is equal to {\rm id}_H. Therefore, at the K-theoretic level, we have that the maps K_0(C(H))\rightarrow K_0(C_\kappa(\mathbb{G})) must be injective, and that is enough to show that the map on the left side is 0.

We use the counit \varepsilon_H which is a unital *-morphism from C(H) to \mathbb{C}. We have that ({\rm id}\otimes \varepsilon_H)\beta = {\rm id}. Therefore the map \varepsilon_{H\ast} is a left-inverse to the map \beta_\ast. Hence:

with the composition being the identity morphism. This implies that [\beta(e_{11})]= \beta_\ast ([e_{11}]) generates a direct factor.

In the case B= M_N(\mathbb{C}), we have the isomorphism M_N(\mathbb{C}) \otimes C(\mathbb{G}) \simeq C(\widetilde{\mathbb{G}}). Therefore we can compute the K-theory groups of C(\widetilde{\mathbb{G}}) only. The K-theory groups of C(\mathbb{Z}_s)\simeq \mathbb{C}^s are

A basis of generators for K_0(\mathbb{C}^s)\simeq \mathbb{Z}^{s} is the classes [e_i] for 1\leq i\leq s but we remark that we can replace the last generator by the class [1]= \sum_{i=1}^s [e_i], so that \mathbb{Z}\cdot [1] is a direct factor in it. We immediately have that

For the K_0 group we use the exact sequence

This gives the isomorphism

By the previous discussions on the generators, we know that \mathbb{Z}[1] and \mathbb{Z}[\beta(e_{11})] are direct factors, and therefore we get that K_0(C(\mathbb{G}))\simeq K_0(C(\widetilde{\mathbb{G}}))\simeq \mathbb{Z}^s \oplus \mathbb{Z}_N.

The proof is exactly the same as the previous case. Using that:

We get that the two copies of \mathbb{Z} in K_0 are identified, and the result.